diff --git a/cuda-oxide/.devcontainer/README.md b/cuda-oxide/.devcontainer/README.md
index 26f3fbd4a5..01bcde0c46 100644
--- a/cuda-oxide/.devcontainer/README.md
+++ b/cuda-oxide/.devcontainer/README.md
@@ -6,7 +6,7 @@ This dev container provides the toolchain expected by cuda-oxide:
 - CUDA Toolkit 13.0
 - LLVM 21 with NVPTX support
 - Clang 21 resource headers for `bindgen`
-- Rust `nightly-2026-08-28` with `rust-src`, `rustc-dev`, `rust-analyzer`,
+- Rust `nightly-2026-10-11` with `rust-src`, `rustc-dev`, `rust-analyzer`,
   `rustfmt` and `clippy`. `llvm-tools` is not preinstalled in the image;
   rustup adds it on your first cargo command under `cuda-oxide/`, because
   `cuda-oxide/rust-toolchain.toml` names it. The image ships LLVM 21 and
diff --git a/cuda-oxide/.devcontainer/devcontainer.json b/cuda-oxide/.devcontainer/devcontainer.json
index 577857720e..06fa7b4ba7 100644
--- a/cuda-oxide/.devcontainer/devcontainer.json
+++ b/cuda-oxide/.devcontainer/devcontainer.json
@@ -17,7 +17,7 @@
       "upgradePackages": false
     },
     "ghcr.io/devcontainers/features/rust:1": {
-      "version": "nightly-2026-08-28",
+      "version": "nightly-2026-10-11",
       "profile": "minimal",
       "components": "rust-src,rustc-dev,rust-analyzer,rustfmt,clippy"
     }
diff --git a/cuda-oxide/README.md b/cuda-oxide/README.md
index 15b3610035..586f936a72 100644
--- a/cuda-oxide/README.md
+++ b/cuda-oxide/README.md
@@ -164,7 +164,7 @@ From `cuda-oxide/`, `cargo oxide` works out of the box via a workspace alias.
 For use outside the repo (your own projects), install it with the pinned nightly toolchain:
 
 ```bash
-cargo +nightly-2026-08-28 install --git https://github.com/NVIDIA/cuda-rust.git cargo-oxide
+cargo +nightly-2026-10-11 install --git https://github.com/NVIDIA/cuda-rust.git cargo-oxide
 ```
 
 On first run, `cargo-oxide` will automatically fetch and build the codegen backend.
@@ -187,8 +187,8 @@ nix run 'github:NVIDIA/cuda-rust?dir=cuda-oxide#new' -- my-project
 ```bash
 # Toolchain installed automatically via rust-toolchain.toml
 # Manual install if needed:
-rustup toolchain install nightly-2026-08-28
-rustup component add rust-src rustc-dev rust-analyzer clippy rustfmt llvm-tools --toolchain nightly-2026-08-28
+rustup toolchain install nightly-2026-10-11
+rustup component add rust-src rustc-dev rust-analyzer clippy rustfmt llvm-tools --toolchain nightly-2026-10-11
 ```
 
 #### CUDA
diff --git a/cuda-oxide/crates/cargo-oxide/README.md b/cuda-oxide/crates/cargo-oxide/README.md
index 0149290f93..ce725327f0 100644
--- a/cuda-oxide/crates/cargo-oxide/README.md
+++ b/cuda-oxide/crates/cargo-oxide/README.md
@@ -13,7 +13,7 @@ Replaces the previous `xtask` pattern with a proper cargo subcommand that works
 Install with the project's pinned nightly toolchain:
 
 ```bash
-cargo +nightly-2026-08-28 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
+cargo +nightly-2026-10-11 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
 ```
 
 On first run, `cargo-oxide` will automatically fetch and build the codegen backend if it's not already available.
diff --git a/cuda-oxide/crates/cargo-oxide/src/commands/scaffold.rs b/cuda-oxide/crates/cargo-oxide/src/commands/scaffold.rs
index d3988d2ec0..599dd86344 100644
--- a/cuda-oxide/crates/cargo-oxide/src/commands/scaffold.rs
+++ b/cuda-oxide/crates/cargo-oxide/src/commands/scaffold.rs
@@ -20,7 +20,7 @@ const GIT_REPO: &str = "https://github.com/NVlabs/cuda-oxide.git";
 pub(super) const SHARED_HOST_CRATES_VERSION: &str = "0.4.0";
 
 const RUST_TOOLCHAIN_TOML: &str = r#"[toolchain]
-channel = "nightly-2026-08-28"
+channel = "nightly-2026-10-11"
 components = ["rust-src", "rustc-dev", "rust-analyzer", "clippy", "rustfmt", "llvm-tools"]
 "#;
 
diff --git a/cuda-oxide/crates/cuda-device/src/atomic.rs b/cuda-oxide/crates/cuda-device/src/atomic.rs
index c39c61cd25..84423472ed 100644
--- a/cuda-oxide/crates/cuda-device/src/atomic.rs
+++ b/cuda-oxide/crates/cuda-device/src/atomic.rs
@@ -490,7 +490,7 @@ define_float_atomic! {
     ///
     /// `swap` lowers to hardware `atom.exch.b32`; this API does not expose
     /// `compare_exchange` for floats. `fetch_add` uses LLVM `atomicrmw fadd`.
-    /// With the pinned `nightly-2026-08-28` toolchain's LLVM 23 NVPTX defaults,
+    /// With the pinned `nightly-2026-10-11` toolchain's LLVM 23 NVPTX defaults,
     /// global and generic `f32` adds use a compare-and-swap loop; shared `f32`
     /// and `f64` adds still use native instructions on supported targets.
     ///
diff --git a/cuda-oxide/crates/cuda-device/src/generated/async_copy.rs b/cuda-oxide/crates/cuda-device/src/generated/async_copy.rs
index e44c9a333c..10e6c3e0b2 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/async_copy.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/async_copy.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::async_copy` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/atomic.rs b/cuda-oxide/crates/cuda-device/src/generated/atomic.rs
index 47a5053d24..425aedc4ac 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/atomic.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::atomic` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/bf16.rs b/cuda-oxide/crates/cuda-device/src/generated/bf16.rs
index fb8215c889..57652a5228 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/bf16.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/bf16.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::bf16` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/bf16x2.rs b/cuda-oxide/crates/cuda-device/src/generated/bf16x2.rs
index b9f4be0b75..aa95140971 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/bf16x2.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/bf16x2.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::bf16x2` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/clc.rs b/cuda-oxide/crates/cuda-device/src/generated/clc.rs
index 94c9cae176..0c9b7ff9aa 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/clc.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::clc` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/cluster_barrier.rs b/cuda-oxide/crates/cuda-device/src/generated/cluster_barrier.rs
index 9115aec1b5..a4b905b524 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/cluster_barrier.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::cluster` to keep public paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/cluster_memory.rs b/cuda-oxide/crates/cuda-device/src/generated/cluster_memory.rs
index d665c6fc59..04d544f30c 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/cluster_memory.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::cluster` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/cluster_sreg.rs b/cuda-oxide/crates/cuda-device/src/generated/cluster_sreg.rs
index 4e2ce888c0..6be477a270 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/cluster_sreg.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/cluster_sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // This file is included inside `cuda_device::cluster`.
 // Its private leaves let public helpers compose generated reads.
diff --git a/cuda-oxide/crates/cuda-device/src/generated/convert.rs b/cuda-oxide/crates/cuda-device/src/generated/convert.rs
index f84862afdd..36a19cb533 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/convert.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/convert.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::convert`.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/counted_barrier.rs b/cuda-oxide/crates/cuda-device/src/generated/counted_barrier.rs
index 877e00f15f..872d91cee2 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/counted_barrier.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/counted_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::barrier` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/debug_control.rs b/cuda-oxide/crates/cuda-device/src/generated/debug_control.rs
index 6f52412e9a..d6828a8aba 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/debug_control.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::debug` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/debug_sreg.rs b/cuda-oxide/crates/cuda-device/src/generated/debug_sreg.rs
index 2028a15d31..ee39d4257d 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/debug_sreg.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/debug_sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::debug` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/dotprod.rs b/cuda-oxide/crates/cuda-device/src/generated/dotprod.rs
index 2e1e16f829..37d8248126 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/dotprod.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::dotprod` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/f16.rs b/cuda-oxide/crates/cuda-device/src/generated/f16.rs
index e0308d047a..2ec7953393 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/f16.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/f16.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::f16` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/f16x2.rs b/cuda-oxide/crates/cuda-device/src/generated/f16x2.rs
index 00085fa0b6..d985923438 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/f16x2.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/f16x2.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::f16x2` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/f32x2.rs b/cuda-oxide/crates/cuda-device/src/generated/f32x2.rs
index c5de9a8539..349e4a4d35 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/f32x2.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/f32x2.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::f32x2` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/fence.rs b/cuda-oxide/crates/cuda-device/src/generated/fence.rs
index 685f7e4c38..88ee2432f3 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/fence.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/fence.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // This file is included inside `cuda_device::fence` so existing paths stay stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/float.rs b/cuda-oxide/crates/cuda-device/src/generated/float.rs
index fbe53bd878..4dcf1edb9b 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/float.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/float.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::float`.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/grid_dependency.rs b/cuda-oxide/crates/cuda-device/src/generated/grid_dependency.rs
index 94d7848ee7..3fb0c6490c 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/grid_dependency.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/grid_dependency.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::grid` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/grid_sreg.rs b/cuda-oxide/crates/cuda-device/src/generated/grid_sreg.rs
index ed3fa06aeb..9b98009d81 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/grid_sreg.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/grid_sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::grid` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/i16x2.rs b/cuda-oxide/crates/cuda-device/src/generated/i16x2.rs
index 054da50cd7..e99a610b45 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/i16x2.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/i16x2.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::i16x2` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/int.rs b/cuda-oxide/crates/cuda-device/src/generated/int.rs
index 87a798b163..3788f17e4b 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/int.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/int.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::int` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/ldmatrix.rs b/cuda-oxide/crates/cuda-device/src/generated/ldmatrix.rs
index b288951019..c90b321c91 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/ldmatrix.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::wmma` to keep public paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/mbarrier_basic.rs b/cuda-oxide/crates/cuda-device/src/generated/mbarrier_basic.rs
index eb377051e6..f7e8c4e54a 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/mbarrier_basic.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::barrier` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/mbarrier_extended.rs b/cuda-oxide/crates/cuda-device/src/generated/mbarrier_extended.rs
index 7787d693cf..0f1714ad05 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/mbarrier_extended.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::barrier` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/movmatrix.rs b/cuda-oxide/crates/cuda-device/src/generated/movmatrix.rs
index 09ccf2ea4d..3b0f283d9b 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/movmatrix.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::wmma` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/prmt.rs b/cuda-oxide/crates/cuda-device/src/generated/prmt.rs
index 30fb07902d..d7a9131ae3 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/prmt.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::prmt` to keep the public paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/register_control.rs b/cuda-oxide/crates/cuda-device/src/generated/register_control.rs
index 91d3e7f239..86e1231142 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/register_control.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/register_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::thread` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/register_mma.rs b/cuda-oxide/crates/cuda-device/src/generated/register_mma.rs
index f623acbe07..e643c970b3 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/register_mma.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/register_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::wmma` to keep public paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/shared_sreg.rs b/cuda-oxide/crates/cuda-device/src/generated/shared_sreg.rs
index 35198aaa6f..037c21f721 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/shared_sreg.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/shared_sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::shared` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/sparse_mma.rs b/cuda-oxide/crates/cuda-device/src/generated/sparse_mma.rs
index a1e8bd0f63..8d0474ed3f 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/sparse_mma.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::wmma` to keep public paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/sreg.rs b/cuda-oxide/crates/cuda-device/src/generated/sreg.rs
index f972a0a046..412e03bf24 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/sreg.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // This file is included lexically inside `cuda_device::thread` so existing
 // DefPaths remain stable during the generated-intrinsics migration.
diff --git a/cuda-oxide/crates/cuda-device/src/generated/stmatrix.rs b/cuda-oxide/crates/cuda-device/src/generated/stmatrix.rs
index 73f22a70d0..ece2a3992a 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/stmatrix.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::tcgen05` to keep public paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/tcgen05.rs b/cuda-oxide/crates/cuda-device/src/generated/tcgen05.rs
index cb1712552e..0ad1f01a09 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/tcgen05.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/tcgen05.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::tcgen05` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/tma.rs b/cuda-oxide/crates/cuda-device/src/generated/tma.rs
index 36156f0b8d..92226b0c88 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/tma.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::tma` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/warp_sreg.rs b/cuda-oxide/crates/cuda-device/src/generated/warp_sreg.rs
index 8019adef7b..b8f8444bac 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/warp_sreg.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/warp_sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::warp` to keep existing paths stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/generated/wgmma_control.rs b/cuda-oxide/crates/cuda-device/src/generated/wgmma_control.rs
index fd1ea8426e..6a734aa7fb 100644
--- a/cuda-oxide/crates/cuda-device/src/generated/wgmma_control.rs
+++ b/cuda-oxide/crates/cuda-device/src/generated/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 // Included inside `cuda_device::wgmma` to keep its public API stable.
 
diff --git a/cuda-oxide/crates/cuda-device/src/vector.rs b/cuda-oxide/crates/cuda-device/src/vector.rs
index 50a7db4a63..edb513e6c3 100644
--- a/cuda-oxide/crates/cuda-device/src/vector.rs
+++ b/cuda-oxide/crates/cuda-device/src/vector.rs
@@ -165,31 +165,52 @@ macro_rules! define_vector {
 
 define_vector!(
     /// Two `f32` in one 64-bit transaction.
-    F32x2, f32, 2, 8
+    F32x2,
+    f32,
+    2,
+    8
 );
 define_vector!(
     /// Four `f32` in one 128-bit transaction. The `float4` equivalent.
-    F32x4, f32, 4, 16
+    F32x4,
+    f32,
+    4,
+    16
 );
 define_vector!(
     /// Two `f64` in one 128-bit transaction.
-    F64x2, f64, 2, 16
+    F64x2,
+    f64,
+    2,
+    16
 );
 define_vector!(
     /// Two `u32` in one 64-bit transaction.
-    U32x2, u32, 2, 8
+    U32x2,
+    u32,
+    2,
+    8
 );
 define_vector!(
     /// Four `u32` in one 128-bit transaction.
-    U32x4, u32, 4, 16
+    U32x4,
+    u32,
+    4,
+    16
 );
 define_vector!(
     /// Four `u16` in one 64-bit transaction.
-    U16x4, u16, 4, 8
+    U16x4,
+    u16,
+    4,
+    8
 );
 define_vector!(
     /// Eight `u16` in one 128-bit transaction. The width packed f16 pairs use.
-    U16x8, u16, 8, 16
+    U16x8,
+    u16,
+    8,
+    16
 );
 
 /// View a flat slice as over-aligned vectors, or `None` if it does not divide.
diff --git a/cuda-oxide/crates/cuda-device/tests/compile_fail/swizzle_non_involutive.stderr b/cuda-oxide/crates/cuda-device/tests/compile_fail/swizzle_non_involutive.stderr
index 599474d478..b5d29acc9d 100644
--- a/cuda-oxide/crates/cuda-device/tests/compile_fail/swizzle_non_involutive.stderr
+++ b/cuda-oxide/crates/cuda-device/tests/compile_fail/swizzle_non_involutive.stderr
@@ -6,6 +6,8 @@ error[E0080]: evaluation panicked: Swizzle<B, M, S> requires M + B <= S, otherwi
   |
  ::: src/swizzle.rs
   |
+  |   impl<const B: usize, const M: usize, const S: usize> Swizzle<B, M, S> {
+  |       /// Compile-time check that the swizzle is an involution.
   |       const INVOLUTION: () = assert!(
   |  ____________________________-
   | |         M + B <= S,
diff --git a/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_runtime_mismatch.stderr b/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_runtime_mismatch.stderr
index 4ca8d8067f..fb4421dbec 100644
--- a/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_runtime_mismatch.stderr
+++ b/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_runtime_mismatch.stderr
@@ -13,3 +13,17 @@ note: method defined here
    |
    |     pub fn get_mut<'kernel>(&mut self, idx: ThreadIndex<'kernel, IndexSpace>) -> Option<&mut T> {
    |            ^^^^^^^
+note: the `get_mut` call is resolved to the method in `cuda_device::DisjointSlice<'a, T, IndexSpace>`, shadowing the method of the same name on the inherent impl for `cuda_device::DisjointSlice<'a, T, cuda_device::Runtime2DIndex>`
+  --> tests/compile_fail/thread_index_runtime_mismatch.rs:18:17
+   |
+18 |     let _ = out.get_mut(idx);
+   |                 ^^^^^^^ refers to `cuda_device::DisjointSlice::<'a, T, IndexSpace>::get_mut`
+   = note: additionally, there is 1 other available method that isn't in scope
+help: you might have meant to call one of the other methods; you can use the fully-qualified path to call one of them explicitly
+   |
+18 -     let _ = out.get_mut(idx);
+18 +     let _ = DisjointSlice::get_mut(&mut out, idx);
+   |
+18 -     let _ = out.get_mut(idx);
+18 +     let _ = <_ as std::slice::SliceIndex<_>>::get_mut(out, idx);
+   |
diff --git a/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_stride_mismatch.stderr b/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_stride_mismatch.stderr
index 7eed21fb06..96a0802bce 100644
--- a/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_stride_mismatch.stderr
+++ b/cuda-oxide/crates/cuda-device/tests/compile_fail/thread_index_stride_mismatch.stderr
@@ -13,3 +13,17 @@ note: method defined here
    |
    |     pub fn get_mut<'kernel>(&mut self, idx: ThreadIndex<'kernel, IndexSpace>) -> Option<&mut T> {
    |            ^^^^^^^
+note: the `get_mut` call is resolved to the method in `cuda_device::DisjointSlice<'a, T, IndexSpace>`, shadowing the method of the same name on the inherent impl for `cuda_device::DisjointSlice<'a, T, cuda_device::Runtime2DIndex>`
+  --> tests/compile_fail/thread_index_stride_mismatch.rs:11:17
+   |
+11 |     let _ = out.get_mut(idx);
+   |                 ^^^^^^^ refers to `cuda_device::DisjointSlice::<'a, T, IndexSpace>::get_mut`
+   = note: additionally, there is 1 other available method that isn't in scope
+help: you might have meant to call one of the other methods; you can use the fully-qualified path to call one of them explicitly
+   |
+11 -     let _ = out.get_mut(idx);
+11 +     let _ = DisjointSlice::get_mut(&mut out, idx);
+   |
+11 -     let _ = out.get_mut(idx);
+11 +     let _ = <_ as std::slice::SliceIndex<_>>::get_mut(out, idx);
+   |
diff --git a/cuda-oxide/crates/cuda-host/src/entry_registry.rs b/cuda-oxide/crates/cuda-host/src/entry_registry.rs
index b12885c3af..657da3159a 100644
--- a/cuda-oxide/crates/cuda-host/src/entry_registry.rs
+++ b/cuda-oxide/crates/cuda-host/src/entry_registry.rs
@@ -195,7 +195,7 @@ fn type_id_divergence_message(
     format!(
         "generic kernel PTX entry `{requested}` was not found, but the loaded module DOES \
          contain `{base}` under a different type-identity hash: {found}. The host \
-         (core::intrinsics::type_id) and the device backend (tcx.type_id_hash) disagreed on \
+         (core::intrinsics::reflection::type_id) and the device backend (tcx.type_id_hash) disagreed on \
          the kernel's `_TID_` name, which means the TypeId-to-u128 extraction in \
          cuda_host::type_id is no longer value-correct on this toolchain. Host and device \
          code must be built by the same pinned nightly; re-verify the TypeId contract in \
@@ -329,7 +329,10 @@ mod tests {
             "{message}"
         );
         // The message must identify the two hash producers and the remedy.
-        assert!(message.contains("core::intrinsics::type_id"), "{message}");
+        assert!(
+            message.contains("core::intrinsics::reflection::type_id"),
+            "{message}"
+        );
         assert!(message.contains("tcx.type_id_hash"), "{message}");
         assert!(message.contains("same pinned nightly"), "{message}");
         assert!(message.contains("compute_kernel_export_name"), "{message}");
diff --git a/cuda-oxide/crates/cuda-host/src/launch.rs b/cuda-oxide/crates/cuda-host/src/launch.rs
index c6033b1dc1..c3a2266731 100644
--- a/cuda-oxide/crates/cuda-host/src/launch.rs
+++ b/cuda-oxide/crates/cuda-host/src/launch.rs
@@ -164,7 +164,7 @@ pub trait CudaKernel {
 /// The bound on `ptx_name` is intentionally just whatever the kernel
 /// itself declared — typically `Copy` on the value-passed generics. No
 /// `'static` is added: the helper `cuda_host::type_id_u128` uses
-/// `core::intrinsics::type_id`, which is bound `T: ?Sized` rather than
+/// `core::intrinsics::reflection::type_id`, which is bound `T: ?Sized` rather than
 /// `T: 'static`, so closures that borrow non-`'static` data still go
 /// through the typed launch path (just like they did under the legacy
 /// `type_name`-based scheme).
diff --git a/cuda-oxide/crates/cuda-host/src/lib.rs b/cuda-oxide/crates/cuda-host/src/lib.rs
index 5e6a866a7a..1fa90924a7 100644
--- a/cuda-oxide/crates/cuda-host/src/lib.rs
+++ b/cuda-oxide/crates/cuda-host/src/lib.rs
@@ -3,7 +3,7 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 
-// Required for `type_id_u128`'s use of `core::intrinsics::type_id`. The
+// Required for `type_id_u128`'s use of `core::intrinsics::reflection::type_id`. The
 // intrinsic is the only way to obtain the same 128-bit hash the backend
 // uses while keeping the bound at `T: ?Sized` (stable `TypeId::of` would
 // force `T: 'static` on every kernel marker — see `type_id.rs` for why
diff --git a/cuda-oxide/crates/cuda-host/src/type_id.rs b/cuda-oxide/crates/cuda-host/src/type_id.rs
index 8a46e85cf1..e9aa917b55 100644
--- a/cuda-oxide/crates/cuda-host/src/type_id.rs
+++ b/cuda-oxide/crates/cuda-host/src/type_id.rs
@@ -10,7 +10,7 @@
 //! API would force a `T: 'static` bound on the kernel marker, which would in
 //! turn reject perfectly valid non-`'static` borrowing closures (e.g. a kernel
 //! launcher capturing `&[f32]` from a stack frame the caller keeps alive
-//! across the launch). The `core::intrinsics::type_id` form has bound
+//! across the launch). The `core::intrinsics::reflection::type_id` form has bound
 //! `T: ?Sized` — i.e. no `'static` requirement — and produces the exact same
 //! 128-bit value that `tcx.type_id_hash` does for that type, because both go
 //! through the same `erase_and_anonymize_regions` + stable-hash pipeline.
@@ -20,7 +20,7 @@
 //! every monomorphized type and const argument, so one hash covers the complete
 //! specialization without inventing a parallel encoding for const values.
 //!
-//! Framing note for future contributors: `core::intrinsics::type_id` is an
+//! Framing note for future contributors: `core::intrinsics::reflection::type_id` is an
 //! internal API and requires `#![feature(core_intrinsics)]` on the owning
 //! crate. cuda-oxide already ships against `rustc_private` and pins a
 //! nightly toolchain, so this is inside our existing risk surface — but the
@@ -54,7 +54,7 @@ static GENERIC_KERNEL_NAMES: OnceLock<Mutex<HashMap<KernelNameKey, &'static str>
 /// caller (the borrow must outlive `stream.synchronize()`).
 #[inline]
 pub fn type_id_u128<T: ?Sized>() -> u128 {
-    let id = const { core::intrinsics::type_id::<T>() };
+    let id = const { core::intrinsics::reflection::type_id::<T>() };
     unsafe { core::mem::transmute::<TypeId, u128>(id) }
 }
 
diff --git a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/fixtures.rs b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/fixtures.rs
index 082bfbf15e..c9bbbdc9ff 100644
--- a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/fixtures.rs
+++ b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/fixtures.rs
@@ -1471,7 +1471,7 @@ pub(super) fn test_wgmma_control_admission() -> WgmmaControlAdmission {
 
 pub(super) fn test_special_register_admission() -> SpecialRegisterAdmission {
     SpecialRegisterAdmission {
-        llvm_evidence_profile: "rust-llvm-23.1.0-16696adc".into(),
+        llvm_evidence_profile: "rust-llvm-23.1.3-aab11ee0".into(),
         libnvvm_evidence_profile: "cuda-13.3-libnvvm-13.3.33".into(),
         runtime_validation: RuntimeValidation::Unexecuted,
         registers: REVIEWED_SPECIAL_REGISTERS.into(),
diff --git a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/misc_families.rs b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/misc_families.rs
index 8f50c158f2..bbe5c2c982 100644
--- a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/misc_families.rs
+++ b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/misc_families.rs
@@ -233,7 +233,7 @@ fn active_debug_control_sources_parse_and_prove_both_backend_routes() {
 
     let evidence = vec![
         read_evidence_file(
-            &repo_root.join("intrinsics/evidence/rust-llvm-23.1.0-16696adc-debug-control.json"),
+            &repo_root.join("intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-debug-control.json"),
         )
         .unwrap(),
         read_evidence_file(
@@ -241,7 +241,7 @@ fn active_debug_control_sources_parse_and_prove_both_backend_routes() {
         )
         .unwrap(),
     ];
-    let indexed = index_evidence(&evidence, "16696adcd119e6ba9cc175207d984d7021211acb").unwrap();
+    let indexed = index_evidence(&evidence, "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019").unwrap();
     for record in &records {
         let routes = resolve_backend_lowerings(record, &indexed).unwrap();
         assert_eq!(routes.len(), 2);
diff --git a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/mma.rs b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/mma.rs
index d62473d0ff..1d05c5e2e7 100644
--- a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/mma.rs
+++ b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/mma.rs
@@ -969,7 +969,7 @@ fn standard_fp8_resolves_exact_routes_and_target_floors() {
         (
             IntrinsicBackend::LlvmNvptx,
             BackendLoweringMechanism::InlinePtx,
-            "rust-llvm-23.1.0-16696adc",
+            "rust-llvm-23.1.3-aab11ee0",
         ),
         (
             IntrinsicBackend::LibNvvm,
@@ -1028,7 +1028,11 @@ fn standard_fp8_resolves_exact_routes_and_target_floors() {
             assert!(llvm_route.stages.iter().any(|stage| {
                 stage.stage == EvidenceStageKind::PtxAssembly
                     && stage.mechanism == Some(BackendLoweringMechanism::InlinePtx)
-                    && stage.outcome == "failed"
+                    && stage.outcome == "not_revalidated"
+                    && stage.detail.contains("Original outcome: failed.")
+                    && stage
+                        .detail
+                        .contains("e457a7b0d326d67b4322ef0d11bd715cfaeda48f")
                     && stage.targets == ["sm_89", "ptx86"]
             }));
         }
diff --git a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/sreg_threadfence.rs b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/sreg_threadfence.rs
index 593cd62229..f03190029c 100644
--- a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/sreg_threadfence.rs
+++ b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/sreg_threadfence.rs
@@ -245,7 +245,7 @@ fn special_register_evidence_validates_both_backend_routes() {
     let policies = expand_special_register_admission(&test_special_register_admission()).unwrap();
     let mut evidence_files = vec![
         read_evidence_file(
-            &repo_root.join("intrinsics/evidence/rust-llvm-23.1.0-16696adc-special-registers.json"),
+            &repo_root.join("intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-special-registers.json"),
         )
         .unwrap(),
         read_evidence_file(
@@ -253,7 +253,7 @@ fn special_register_evidence_validates_both_backend_routes() {
         )
         .unwrap(),
     ];
-    let llvm_revision = "16696adcd119e6ba9cc175207d984d7021211acb";
+    let llvm_revision = "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019";
     let indexed = index_evidence(&evidence_files, llvm_revision).unwrap();
     for policy in &policies {
         for lowering in &policy.backend_lowerings {
diff --git a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/tma_clc_cluster.rs b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/tma_clc_cluster.rs
index ceb416cb46..c876fea84a 100644
--- a/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/tma_clc_cluster.rs
+++ b/cuda-oxide/crates/cuda-intrinsics-gen/src/resolve/tests/tma_clc_cluster.rs
@@ -632,12 +632,12 @@ fn compact_cluster_barrier_admission_and_semantics_fail_closed() {
 fn cluster_barrier_evidence_validates_both_backend_routes() {
     let repo_root = Path::new(env!("CARGO_MANIFEST_DIR")).join("../..");
     let mut admission = test_cluster_barrier_admission();
-    admission.llvm_evidence_profile = "rust-llvm-23.1.0-16696adc".into();
+    admission.llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0".into();
     admission.libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-cluster-barrier".into();
     let policies = expand_cluster_barrier_admission(&admission).unwrap();
     let evidence_files = vec![
         read_evidence_file(
-            &repo_root.join("intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-barrier.json"),
+            &repo_root.join("intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-barrier.json"),
         )
         .unwrap(),
         read_evidence_file(
@@ -646,7 +646,7 @@ fn cluster_barrier_evidence_validates_both_backend_routes() {
         .unwrap(),
     ];
     let indexed =
-        index_evidence(&evidence_files, "16696adcd119e6ba9cc175207d984d7021211acb").unwrap();
+        index_evidence(&evidence_files, "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019").unwrap();
 
     for policy in &policies {
         for lowering in &policy.backend_lowerings {
@@ -670,7 +670,7 @@ fn cluster_barrier_evidence_validates_both_backend_routes() {
     }
     let indexed = index_evidence(
         &missing_typed_failure,
-        "16696adcd119e6ba9cc175207d984d7021211acb",
+        "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
     )
     .unwrap();
     let policy = &policies[0];
diff --git a/cuda-oxide/crates/cuda-intrinsics/src/generated/abi_v1.rs b/cuda-oxide/crates/cuda-intrinsics/src/generated/abi_v1.rs
index ce0289c0dd..98d67d5b98 100644
--- a/cuda-oxide/crates/cuda-intrinsics/src/generated/abi_v1.rs
+++ b/cuda-oxide/crates/cuda-intrinsics/src/generated/abi_v1.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Raw ABI functions recognized by cuda-oxide.
 
diff --git a/cuda-oxide/crates/cuda-intrinsics/src/generated/mod.rs b/cuda-oxide/crates/cuda-intrinsics/src/generated/mod.rs
index 3ab1989e8d..fb3435f9b8 100644
--- a/cuda-oxide/crates/cuda-intrinsics/src/generated/mod.rs
+++ b/cuda-oxide/crates/cuda-intrinsics/src/generated/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 #[doc(hidden)]
 #[path = "abi_v1.rs"]
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_fakes_row_width.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_fakes_row_width.stderr
index a4cedec322..6a068a9dd8 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_fakes_row_width.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_fakes_row_width.stderr
@@ -12,5 +12,4 @@ note: required by a bound in `LoadedModule::alias_fakes_row_width`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::alias_fakes_row_width`
 ...
 22 |     pub fn alias_fakes_row_width(mut out: DisjointSlice<f32, Runtime2DIndex>) {
-   |            --------------------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_hides_row_width.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_hides_row_width.stderr
index 4bc77dcf68..1e2dd6faa7 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_hides_row_width.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_alias_hides_row_width.stderr
@@ -12,5 +12,4 @@ note: required by a bound in `LoadedModule::alias_hides_row_width`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::alias_hides_row_width`
 ...
 22 |     pub fn alias_hides_row_width(mut out: DisjointSlice<f32, Rt>) {
-   |            --------------------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_fake_disjoint_slice.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_fake_disjoint_slice.stderr
index 530af8ce95..a0d8471d81 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_fake_disjoint_slice.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_fake_disjoint_slice.stderr
@@ -26,7 +26,6 @@ note: required by a bound in `LoadedModule::prepare_lookalike`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::prepare_lookalike`
 ...
 21 |     pub fn lookalike(mut out: DisjointSlice<u32>) {
-   |            --------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
 
 error[E0277]: the trait bound `for<'__cuda_oxide_disjoint> DisjointSlice<'__cuda_oxide_disjoint, u32>: cuda_device::disjoint::__LaunchContractDisjointSliceAbi<u32, false>` is not satisfied
@@ -57,5 +56,4 @@ note: required by a bound in `LoadedModule::prepare_lookalike`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::prepare_lookalike`
 ...
 21 |     pub fn lookalike(mut out: DisjointSlice<u32>) {
-   |            --------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_misleading_index_alias.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_misleading_index_alias.stderr
index 50bc23cd6f..3208581be6 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_misleading_index_alias.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_misleading_index_alias.stderr
@@ -22,5 +22,4 @@ note: required by a bound in `LoadedModule::prepare_misleading_name`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::prepare_misleading_name`
 ...
 13 |     pub fn misleading_name(mut out: DisjointSlice<u32, Index2D>) {
-   |            --------------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_reordered_disjoint_alias.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_reordered_disjoint_alias.stderr
index 6509665d5d..851917094c 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_reordered_disjoint_alias.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/launch_contract_reordered_disjoint_alias.stderr
@@ -22,7 +22,6 @@ note: required by a bound in `LoadedModule::prepare_reordered`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::prepare_reordered`
 ...
 17 |     pub fn reordered(mut out: DisjointSlice<Index1D, u64>) {
-   |            --------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
 
 error[E0277]: the trait bound `for<'__cuda_oxide_disjoint> cuda_device::DisjointSlice<'__cuda_oxide_disjoint, u64>: cuda_device::disjoint::__LaunchContractDisjointSliceAbi<cuda_device::Index1D, false>` is not satisfied
@@ -49,5 +48,4 @@ note: required by a bound in `LoadedModule::prepare_reordered`
    | ^^^^^^^^^^^^^^ required by this bound in `LoadedModule::prepare_reordered`
 ...
 17 |     pub fn reordered(mut out: DisjointSlice<Index1D, u64>) {
-   |            --------- required by a bound in this associated function
    = note: this error originates in the attribute macro `cuda_module` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_invalid_unroll.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_invalid_unroll.stderr
index a5f714062b..3d37cf49ca 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_invalid_unroll.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_invalid_unroll.stderr
@@ -1,6 +1,7 @@
 error[E0080]: evaluation panicked: partial unroll factor must be in 2..=1024, or 0 for full unrolling
  --> $WORKSPACE/crates/cuda-device/src/thread.rs
   |
+  | pub fn __unroll_config<const FACTOR: u32>() {
   |     const { validate_unroll_factor(FACTOR) }
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ evaluation of `cuda_device::__unroll_config::<1>::{constant#0}` failed inside this call
   |
diff --git a/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_zero_threads.stderr b/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_zero_threads.stderr
index 9d71df7e96..b6d60f72e7 100644
--- a/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_zero_threads.stderr
+++ b/cuda-oxide/crates/cuda-macros/tests/compile_fail/policy_config_zero_threads.stderr
@@ -1,6 +1,7 @@
 error[E0080]: evaluation panicked: launch_bounds maximum threads must be greater than zero
  --> $WORKSPACE/crates/cuda-device/src/thread.rs
   |
+  | pub fn __launch_bounds_config<const MAX_THREADS: u32, const MIN_BLOCKS: u32>() {
   |     const { validate_launch_bounds(MAX_THREADS) }
   |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ evaluation of `cuda_device::__launch_bounds_config::<0, 0>::{constant#0}` failed inside this call
   |
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/active_mask.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/active_mask.rs
index f0cc44758a..6dfd69169a 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/active_mask.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/active_mask.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `active_mask` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/clc.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/clc.rs
index 9a62dd9586..aadf572896 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/clc.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `clc` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_barrier.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_barrier.rs
index aa284eb54e..fe36a8aff8 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_barrier.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `cluster_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_memory.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_memory.rs
index a3e2c44a3b..1e1509cef6 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_memory.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `cluster_memory` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cp_async.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cp_async.rs
index 8955f8cc8c..ead108a430 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cp_async.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/cp_async.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `cp_async` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/debug_control.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/debug_control.rs
index b50a23afae..8cf09104c6 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/debug_control.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `debug_control` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/dotprod.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/dotprod.rs
index f73fef865f..7020044fa2 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/dotprod.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `dotprod` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/elect.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/elect.rs
index 0ada366dc9..72c7864a4f 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/elect.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/elect.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `elect` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/execution_control.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/execution_control.rs
index 52c9ffdd8d..1953d9b216 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/execution_control.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/execution_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `execution_control` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/extended_minmax.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/extended_minmax.rs
index 735115d8a9..fcb91feaf8 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/extended_minmax.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/extended_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `extended_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/integer_minmax.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/integer_minmax.rs
index dde0b8cefd..89d677a028 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/integer_minmax.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/integer_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `integer_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/ldmatrix.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/ldmatrix.rs
index 4f3de70be4..c6c7586c8c 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/ldmatrix.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `ldmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_basic.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_basic.rs
index 31cc8f6be6..3e1feecc57 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_basic.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `mbarrier_basic` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_extended.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_extended.rs
index 6c9be5ac8a..5b15bf968e 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_extended.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `mbarrier_extended` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mod.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mod.rs
index c6aa47c031..fb9ad47daa 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mod.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated target requirements and separately imported LLVM/selection facts.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/movmatrix.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/movmatrix.rs
index ea446d4825..9530f36fbc 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/movmatrix.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `movmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_alu.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_alu.rs
index 74aa24dacd..9dca2e8bfa 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_alu.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_alu.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `packed_alu` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_atomic.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_atomic.rs
index c7e5547922..76f6352d37 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_atomic.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `packed_atomic` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_conversion.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_conversion.rs
index 1fd06948b5..41849141b1 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_conversion.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/packed_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `packed_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/prmt.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/prmt.rs
index 4c4a2f8287..26829231f7 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/prmt.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `prmt` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/redux.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/redux.rs
index bdceaba496..643b0122d8 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/redux.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/redux.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `redux` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/register_mma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/register_mma.rs
index 2ccb0c0db1..d21b5037bf 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/register_mma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/register_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `register_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_arithmetic.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_arithmetic.rs
index 102a59e264..f714ace2f2 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_arithmetic.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_arithmetic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `scalar_arithmetic` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_conversion.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_conversion.rs
index 66bef39204..26c2b9101e 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_conversion.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `scalar_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_math.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_math.rs
index c0b78a50bf..ca977feaf9 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_math.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/scalar_math.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `scalar_math` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sparse_mma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sparse_mma.rs
index 712fbc5a18..60066e6a95 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sparse_mma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `sparse_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sreg.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sreg.rs
index 8ac059d9cc..ced2ff35e1 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sreg.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `sreg` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/stmatrix.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/stmatrix.rs
index 556a7e788e..6987c58ffe 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/stmatrix.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `stmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sync.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sync.rs
index c58ca91aab..e6589fd47f 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sync.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/sync.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `sync` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_cp.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_cp.rs
index e667da42b1..e6dfa4e62d 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_cp.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_cp.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tcgen05_cp` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_ld.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_ld.rs
index f9a55b392c..a71b4e90ce 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_ld.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_ld.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tcgen05_ld` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma.rs
index ab96fe16aa..2c5b7977ea 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tcgen05_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma_ws.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma_ws.rs
index edd04c7507..7519ac479c 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma_ws.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_mma_ws.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tcgen05_mma_ws` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_other.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_other.rs
index fa7086dcc6..8101ef71f2 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_other.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_other.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tcgen05_other` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_st.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_st.rs
index a00308f306..942b0d8d20 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_st.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tcgen05_st.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tcgen05_st` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/active_mask.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/active_mask.rs
index 2fd40c18ed..b4388a4208 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/active_mask.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/active_mask.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `active_mask` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/clc.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/clc.rs
index 6076281fa9..b05ce7cbf9 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/clc.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `clc` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_barrier.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_barrier.rs
index cc0926521e..d13cbf86fb 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_barrier.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `cluster_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_memory.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_memory.rs
index 4635787a76..a35c18829c 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_memory.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `cluster_memory` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cp_async.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cp_async.rs
index 4c2a20c02a..4ea27e9777 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cp_async.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/cp_async.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `cp_async` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/debug_control.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/debug_control.rs
index 123f15d932..be7f8fbead 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/debug_control.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `debug_control` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/dotprod.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/dotprod.rs
index 8081db240d..e56d77b272 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/dotprod.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `dotprod` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/elect.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/elect.rs
index 169baef8cf..c681aa4b93 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/elect.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/elect.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `elect` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/execution_control.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/execution_control.rs
index 78ab2245bb..458a3d4b8f 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/execution_control.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/execution_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `execution_control` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/extended_minmax.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/extended_minmax.rs
index faf89ff5ba..f1503abbd0 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/extended_minmax.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/extended_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `extended_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/integer_minmax.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/integer_minmax.rs
index 053f68fc88..9f808f3507 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/integer_minmax.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/integer_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `integer_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/ldmatrix.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/ldmatrix.rs
index 675e1ca2cf..e6d33d9513 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/ldmatrix.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `ldmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_basic.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_basic.rs
index 5e4809b884..678bbd7350 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_basic.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `mbarrier_basic` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_extended.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_extended.rs
index b25414b0a5..971f157365 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_extended.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `mbarrier_extended` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mod.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mod.rs
index f34109e6e6..eacc5c5122 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mod.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests, grouped per family shard.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/movmatrix.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/movmatrix.rs
index 66b12afa5e..f0e0a4f733 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/movmatrix.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `movmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_alu.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_alu.rs
index 75ff1a863c..0945d8c1f6 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_alu.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_alu.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `packed_alu` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_atomic.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_atomic.rs
index d8ceed3973..7ed4a44ef6 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_atomic.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `packed_atomic` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_conversion.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_conversion.rs
index 6d7fad75c4..6f86271313 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_conversion.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/packed_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `packed_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/prmt.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/prmt.rs
index 214502fbda..491b841ab5 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/prmt.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `prmt` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/redux.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/redux.rs
index c4857e8d56..8647954b1c 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/redux.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/redux.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `redux` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_float.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_float.rs
index 48b805f004..9865a8fa47 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_float.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_float.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `register_mma_float` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_s32.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_s32.rs
index 5216dc659d..478197ec78 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_s32.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/register_mma_s32.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `register_mma_s32` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_arithmetic.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_arithmetic.rs
index ca62c3c7de..60e4b4a4f0 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_arithmetic.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_arithmetic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `scalar_arithmetic` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_conversion.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_conversion.rs
index 98af462f88..d775d2d9c0 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_conversion.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `scalar_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_math.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_math.rs
index 570b8b0232..da31b53b2c 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_math.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/scalar_math.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `scalar_math` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sparse_mma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sparse_mma.rs
index bad07267a2..b2477c9d74 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sparse_mma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `sparse_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sreg.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sreg.rs
index b800b7b44c..8a1758029b 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sreg.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `sreg` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/stmatrix.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/stmatrix.rs
index aa3027531b..c136d5c526 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/stmatrix.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `stmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sync.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sync.rs
index fc5ddff329..aa7f15a2bb 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sync.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/sync.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `sync` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_cp.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_cp.rs
index 3dbb77ebd1..0659c6ce7e 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_cp.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_cp.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tcgen05_cp` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_ld.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_ld.rs
index b01dcd7de4..67318df319 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_ld.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_ld.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tcgen05_ld` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma.rs
index 90092d1fb8..322035b874 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tcgen05_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma_ws.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma_ws.rs
index 589e374de8..1aecf409b0 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma_ws.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_mma_ws.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tcgen05_mma_ws` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_other.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_other.rs
index e2ea3a9ddc..5a2af67e32 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_other.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_other.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tcgen05_other` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_st.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_st.rs
index a324705f5e..0d5606e517 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_st.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tcgen05_st.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tcgen05_st` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tma.rs
index cf0d62e8d4..764bc401f0 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `tma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/vote.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/vote.rs
index 34ab9c2e50..04c7f5e3a7 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/vote.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/vote.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `vote` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_barrier.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_barrier.rs
index e0c0274b88..ce2d1695bb 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_barrier.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `warp_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_match.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_match.rs
index 0a7b559a3b..f20f4c4491 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_match.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_match.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `warp_match` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_shuffle.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_shuffle.rs
index 17608cd3ac..36db24882a 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_shuffle.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/warp_shuffle.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `warp_shuffle` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/wgmma_control.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/wgmma_control.rs
index 2fe8dd720f..447579f72e 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/wgmma_control.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tests/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target tests: `wgmma_control` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tma.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tma.rs
index c0323beb8d..129f12b861 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tma.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `tma` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/vote.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/vote.rs
index 73d122c8b5..033064166b 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/vote.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/vote.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `vote` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_barrier.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_barrier.rs
index b6829d7a5a..d45478d0ce 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_barrier.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `warp_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_match.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_match.rs
index 8517fd196b..7e1c4672ee 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_match.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_match.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `warp_match` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_shuffle.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_shuffle.rs
index 20cb4a2df6..f44e7d1502 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_shuffle.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/warp_shuffle.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `warp_shuffle` intrinsics.
 
diff --git a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/wgmma_control.rs b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/wgmma_control.rs
index e06d161614..90d30ad393 100644
--- a/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/wgmma_control.rs
+++ b/cuda-oxide/crates/cuda-oxide-codegen/src/generated_intrinsic_targets/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated intrinsic target records: `wgmma_control` intrinsics.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/active_mask.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/active_mask.rs
index 449ea2c5e7..a7a20522ab 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/active_mask.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/active_mask.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operation for the generated active warp mask.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/clc.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/clc.rs
index 1c52191590..9a395de71b 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/clc.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated Cluster Launch Control operations.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_barrier.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_barrier.rs
index fe423c1171..7c88557b22 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_barrier.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated operations for the closed cluster-barrier family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_memory.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_memory.rs
index 488275d4d4..a962640b88 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_memory.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for cluster address mapping and remote shared reads.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cp_async.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cp_async.rs
index bfb3bed78c..d4a8b03f03 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cp_async.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/cp_async.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for classic `cp.async` instructions.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/debug_control.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/debug_control.rs
index d82e88c6aa..49d0d28a82 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/debug_control.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for generated PTX debug controls.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/dotprod.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/dotprod.rs
index ae2ab10cf8..a3c1d34d3d 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/dotprod.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for generated packed integer dot products.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/elect.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/elect.rs
index 666b1b6c3f..7d6b492462 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/elect.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/elect.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated warp leader-election operation.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/execution_control.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/execution_control.rs
index ec29e13a32..45be3156b8 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/execution_control.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/execution_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated counted-barrier, grid-dependency, and register-control operations.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/extended_minmax.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/extended_minmax.rs
index 693b61f9c3..41022439a0 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/extended_minmax.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/extended_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for extended floating-point min/max.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/integer_minmax.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/integer_minmax.rs
index 93e4b44632..51892d15b6 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/integer_minmax.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/integer_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for generated extended integer min/max.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/ldmatrix.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/ldmatrix.rs
index fac1180302..1d632153a8 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/ldmatrix.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operation and compatibility carriers for generated `ldmatrix`.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_basic.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_basic.rs
index 38578d4237..8fb1cc51bd 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_basic.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for the basic shared-memory mbarrier lifecycle.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_extended.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_extended.rs
index ec97c54ac3..7ffd7788f6 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_extended.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for extended mbarrier and async-proxy instructions.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mod.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mod.rs
index 3c2463da32..6346816724 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mod.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 mod active_mask;
 mod clc;
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/movmatrix.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/movmatrix.rs
index e7a3136723..f132dfc637 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/movmatrix.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operation for generated in-register matrix transpose.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_alu.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_alu.rs
index 30c46117b3..847de809a9 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_alu.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_alu.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for generated packed floating-point arithmetic.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_atomic.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_atomic.rs
index b7ac0ce129..009cbae284 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_atomic.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for the closed generated packed-atomic family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_conversion.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_conversion.rs
index d992cb6310..ba51dd6199 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_conversion.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/packed_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operation for generated packed conversion.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/prmt.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/prmt.rs
index ea9a88d3ac..d50f35611b 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/prmt.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for the closed generated `prmt` family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/redux.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/redux.rs
index 2236ae8a42..c7b975313e 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/redux.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/redux.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural NVVM operations for the closed generated `redux.sync` family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/register_mma.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/register_mma.rs
index 552a0cb3c3..1ce518c2d3 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/register_mma.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/register_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for generated register-only `mma.sync` variants.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_arithmetic.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_arithmetic.rs
index 726a2256e7..b181effc4d 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_arithmetic.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_arithmetic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for generated scalar arithmetic.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_conversion.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_conversion.rs
index 1d2c361ae5..15b748efa1 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_conversion.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for generated `f32` to TF32 conversions.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_math.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_math.rs
index 424fd246dc..b0119698f4 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_math.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/scalar_math.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for generated scalar math.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sparse_mma.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sparse_mma.rs
index 578e3a3e4e..034c3269a5 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sparse_mma.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! One structural operation for generated sparse MMA variants.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sreg.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sreg.rs
index ed4f1c62d1..47b3ecbf14 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sreg.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural NVVM operations for generated special-register reads.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/stmatrix.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/stmatrix.rs
index c3e587573f..92f23e0c6b 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/stmatrix.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for the four generated `stmatrix` stores.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sync.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sync.rs
index 02d2db9500..90a3220586 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sync.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/sync.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operation for generated CTA synchronization.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tcgen05.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tcgen05.rs
index 990baf92b6..49b72bdabe 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tcgen05.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tcgen05.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated Tensor Core Generation 5 operations.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tma.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tma.rs
index b6457435ee..ea10d67f42 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tma.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated Tensor Memory Accelerator operations.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/vote.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/vote.rs
index def9f11d20..ddec56cc48 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/vote.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/vote.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for the generated `vote.sync` family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_barrier.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_barrier.rs
index 32aab1739b..b46c79b9ca 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_barrier.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operation for generated warp synchronization.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_match.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_match.rs
index c0c01f5ee1..0c7b64afad 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_match.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_match.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for the generated `match.sync` family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_shuffle.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_shuffle.rs
index 767391efc9..099ef58305 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_shuffle.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/warp_shuffle.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for the generated `shfl.sync` family.
 
diff --git a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/wgmma_control.rs b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/wgmma_control.rs
index 9872b0e4b9..0c82363e67 100644
--- a/cuda-oxide/crates/dialect-nvvm/src/ops/generated/wgmma_control.rs
+++ b/cuda-oxide/crates/dialect-nvvm/src/ops/generated/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Structural operations for generated WGMMA controls.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/float_math.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/float_math.rs
index e77e0c21e1..af29e44872 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/float_math.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/float_math.rs
@@ -174,6 +174,58 @@ pub enum RustFloatMathIntrinsic {
 }
 
 impl RustFloatMathIntrinsic {
+    /// Resolve both the historical width-specific paths and the generic unary
+    /// intrinsics introduced by the pinned nightly. Generic calls carry their
+    /// width in rustc's monomorphized type arguments, never in a display name.
+    pub fn from_core_path_with_substs(
+        name: &str,
+        type_substs: &[rustc_public::ty::Ty],
+    ) -> Result<Option<Self>, String> {
+        use rustc_public::ty::{RigidTy, TyKind};
+
+        if let Some(intrinsic) = Self::from_core_path(name) {
+            return Ok(Some(intrinsic));
+        }
+        let float_ty = match type_substs {
+            [ty] => match ty.kind() {
+                TyKind::RigidTy(RigidTy::Float(float_ty)) => Some(float_ty),
+                _ => None,
+            },
+            _ => None,
+        };
+        Self::from_generic_core_path(name, float_ty)
+    }
+
+    fn from_generic_core_path(
+        name: &str,
+        float_ty: Option<rustc_public::ty::FloatTy>,
+    ) -> Result<Option<Self>, String> {
+        use rustc_public::ty::FloatTy;
+
+        // These exact declarations are `fn<T: FloatPrimitive>(T) -> T`.
+        // Reuse the existing f32/f64 placeholders and libdevice lowerings;
+        // unsupported widths must not be silently narrowed.
+        let (f32_intrinsic, f64_intrinsic) = match name {
+            "core::intrinsics::sin" | "std::intrinsics::sin" => (Self::SinF32, Self::SinF64),
+            "core::intrinsics::cos" | "std::intrinsics::cos" => (Self::CosF32, Self::CosF64),
+            "core::intrinsics::exp" | "std::intrinsics::exp" => (Self::ExpF32, Self::ExpF64),
+            "core::intrinsics::exp2" | "std::intrinsics::exp2" => (Self::Exp2F32, Self::Exp2F64),
+            "core::intrinsics::log" | "std::intrinsics::log" => (Self::LogF32, Self::LogF64),
+            "core::intrinsics::log2" | "std::intrinsics::log2" => (Self::Log2F32, Self::Log2F64),
+            "core::intrinsics::log10" | "std::intrinsics::log10" => {
+                (Self::Log10F32, Self::Log10F64)
+            }
+            _ => return Ok(None),
+        };
+        match float_ty {
+            Some(FloatTy::F32) => Ok(Some(f32_intrinsic)),
+            Some(FloatTy::F64) => Ok(Some(f64_intrinsic)),
+            _ => Err(format!(
+                "generic float intrinsic `{name}` requires one f32 or f64 type argument; found {float_ty:?}"
+            )),
+        }
+    }
+
     /// Recognize the libcore intrinsic path that survived into MIR.
     pub fn from_core_path(name: &str) -> Option<Self> {
         match name {
@@ -685,6 +737,61 @@ mod tests {
     use super::*;
     use dialect_mir::rust_intrinsics;
 
+    #[test]
+    fn generic_float_intrinsics_resolve_only_the_declared_float_type() {
+        use RustFloatMathIntrinsic as I;
+        use rustc_public::ty::FloatTy;
+        for (name, f32_variant, f64_variant) in [
+            ("sin", I::SinF32, I::SinF64),
+            ("cos", I::CosF32, I::CosF64),
+            ("exp", I::ExpF32, I::ExpF64),
+            ("exp2", I::Exp2F32, I::Exp2F64),
+            ("log", I::LogF32, I::LogF64),
+            ("log2", I::Log2F32, I::Log2F64),
+            ("log10", I::Log10F32, I::Log10F64),
+        ] {
+            for root in ["core", "std"] {
+                let path = format!("{root}::intrinsics::{name}");
+                assert_eq!(
+                    I::from_generic_core_path(&path, Some(FloatTy::F32)).unwrap(),
+                    Some(f32_variant)
+                );
+                assert_eq!(
+                    I::from_generic_core_path(&path, Some(FloatTy::F64)).unwrap(),
+                    Some(f64_variant)
+                );
+                for unsupported in [None, Some(FloatTy::F16), Some(FloatTy::F128)] {
+                    let error = I::from_generic_core_path(&path, unsupported).unwrap_err();
+                    assert!(
+                        error.contains("requires one f32 or f64 type argument"),
+                        "{error}"
+                    );
+                }
+            }
+        }
+        for path in [
+            "user::intrinsics::exp",
+            "user::core::intrinsics::sin",
+            "core::intrinsics::exp_m1",
+            "core::intrinsics::sqrt",
+            "core::intrinsics::expf32",
+        ] {
+            assert_eq!(
+                I::from_generic_core_path(path, Some(FloatTy::F32)).unwrap(),
+                None,
+                "{path}"
+            );
+        }
+        assert_eq!(
+            I::from_core_path("core::intrinsics::expf32"),
+            Some(I::ExpF32)
+        );
+        assert_eq!(
+            I::from_core_path("core::intrinsics::expf64"),
+            Some(I::ExpF64)
+        );
+    }
+
     /// `f32::max` / `f64::max` / `f32::min` / `f64::min` all lower to the
     /// `_nsz` flavor of the rustc maxNum/minNum intrinsics. Lock the four
     /// `core::intrinsics::*` paths and their `std::intrinsics::*` aliases to
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/active_mask.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/active_mask.rs
index 7363584ce2..c781eedd18 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/active_mask.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/active_mask.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `active_mask` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/clc.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/clc.rs
index 54643ce57c..5a933334a5 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/clc.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `clc` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_barrier.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_barrier.rs
index ea162f2edc..d2440d38bb 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_barrier.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `cluster_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_memory.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_memory.rs
index 695d3c8f6c..3bd583d949 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_memory.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `cluster_memory` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cp_async.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cp_async.rs
index 017ed0c92e..f44f644445 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cp_async.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/cp_async.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `cp_async` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/debug_control.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/debug_control.rs
index 2bd7ca2bae..3e0626b0db 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/debug_control.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `debug_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/dotprod.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/dotprod.rs
index 14363fb24f..8d0a7aa503 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/dotprod.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `dotprod` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/elect.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/elect.rs
index 7d585a7a52..7dfc1e065e 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/elect.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/elect.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `elect` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/execution_control.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/execution_control.rs
index 697372cc9e..29ec4ddbba 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/execution_control.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/execution_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `execution_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/extended_minmax.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/extended_minmax.rs
index 07e0c4f542..3cb746ad02 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/extended_minmax.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/extended_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `extended_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/helpers.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/helpers.rs
index 7e76c72bb5..8b554dd543 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/helpers.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/helpers.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Shared glue for the generated intrinsic dispatch shards.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/integer_minmax.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/integer_minmax.rs
index 62a9d968a9..38ebad84e3 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/integer_minmax.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/integer_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `integer_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/ldmatrix.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/ldmatrix.rs
index cc0df48deb..adadeb2070 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/ldmatrix.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `ldmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_basic.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_basic.rs
index 6a719bb8a1..a1aac30f33 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_basic.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `mbarrier_basic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_extended.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_extended.rs
index 48657a9535..e9b5d0b1d6 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_extended.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `mbarrier_extended` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mod.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mod.rs
index c1370c9621..531e5c92af 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mod.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility path dispatch for CUDA intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/movmatrix.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/movmatrix.rs
index 271dc99c59..040d607946 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/movmatrix.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `movmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_alu.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_alu.rs
index 5f22170a73..93072a0699 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_alu.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_alu.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `packed_alu` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_atomic.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_atomic.rs
index 95ae3a7713..e9bcbe1484 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_atomic.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `packed_atomic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_conversion.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_conversion.rs
index 811e9681b3..c2fc7fee76 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_conversion.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/packed_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `packed_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/prmt.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/prmt.rs
index 47f3018ae3..c8875a535f 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/prmt.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `prmt` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/redux.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/redux.rs
index acb9d2d2be..356348b1e1 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/redux.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/redux.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `redux` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/register_mma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/register_mma.rs
index 2d333957b1..f3df935e0a 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/register_mma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/register_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `register_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_arithmetic.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_arithmetic.rs
index 2593523055..1f887d4ecc 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_arithmetic.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_arithmetic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `scalar_arithmetic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_conversion.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_conversion.rs
index e91e9a70f9..a0c23d4ffe 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_conversion.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `scalar_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_math.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_math.rs
index e33d6cec27..4b6f53ffb0 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_math.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/scalar_math.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `scalar_math` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sparse_mma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sparse_mma.rs
index 9ad19f8a62..cddbd646fe 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sparse_mma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `sparse_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sreg.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sreg.rs
index 0817a6c2da..042d22fe9a 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sreg.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `sreg` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/stmatrix.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/stmatrix.rs
index 6789f53f39..3a5943e341 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/stmatrix.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `stmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sync.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sync.rs
index e0d6949b1a..5c1dbb7527 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sync.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/sync.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `sync` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_cp.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_cp.rs
index 1f4f183f86..44c59ca52d 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_cp.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_cp.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `tcgen05_cp` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_ld.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_ld.rs
index ee83464604..7bb58a39af 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_ld.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_ld.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `tcgen05_ld` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_mma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_mma.rs
index 3c4cff1501..08d6efc001 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_mma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `tcgen05_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_other.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_other.rs
index 02cf1c9317..f21dc2086d 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_other.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_other.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `tcgen05_other` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_st.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_st.rs
index e7123a1cbb..6edf75bcf9 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_st.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tcgen05_st.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `tcgen05_st` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/active_mask.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/active_mask.rs
index 299cdaee53..c4ab0cc802 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/active_mask.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/active_mask.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `active_mask` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/clc.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/clc.rs
index 04e44ae65b..84bf97d723 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/clc.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `clc` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_barrier.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_barrier.rs
index 522e70ad78..9f0bcb6654 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_barrier.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `cluster_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_memory.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_memory.rs
index b1525803a9..b0a5c03ee5 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_memory.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `cluster_memory` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cp_async.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cp_async.rs
index fab4c9c642..75edecc7e8 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cp_async.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/cp_async.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `cp_async` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/debug_control.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/debug_control.rs
index 365e539146..c0b0a46333 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/debug_control.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `debug_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/dotprod.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/dotprod.rs
index 57bca785f5..ebf3d5ae86 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/dotprod.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `dotprod` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/elect.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/elect.rs
index 63a533fa45..4f2a4b0dba 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/elect.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/elect.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `elect` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/execution_control.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/execution_control.rs
index 7b3ca56b96..bfe29b8e9a 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/execution_control.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/execution_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `execution_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/extended_minmax.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/extended_minmax.rs
index 01a5ccca63..fc08e24bd9 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/extended_minmax.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/extended_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `extended_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/integer_minmax.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/integer_minmax.rs
index 9e25c77525..4e168a9be0 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/integer_minmax.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/integer_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `integer_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/ldmatrix.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/ldmatrix.rs
index 2cdb7b9a67..bcd1ae8633 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/ldmatrix.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `ldmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_basic.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_basic.rs
index 569197d5fa..4c4e7a84fc 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_basic.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `mbarrier_basic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_extended.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_extended.rs
index 0670447d85..2cfdad1ac9 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_extended.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `mbarrier_extended` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mod.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mod.rs
index cc78b6207b..773f7f2f0c 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mod.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests, grouped per intrinsic family.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/movmatrix.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/movmatrix.rs
index d52350e08f..dada114a26 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/movmatrix.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `movmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_alu.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_alu.rs
index 453444a7f5..41d58a16a0 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_alu.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_alu.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `packed_alu` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_atomic.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_atomic.rs
index c6c75b16f3..c0c0d792ba 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_atomic.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `packed_atomic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_conversion.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_conversion.rs
index 04c8091720..7c515139fa 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_conversion.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/packed_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `packed_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/prmt.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/prmt.rs
index df75e02029..e614d48ac4 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/prmt.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `prmt` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/redux.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/redux.rs
index 3963fd8c5e..3539b87869 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/redux.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/redux.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `redux` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/register_mma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/register_mma.rs
index 2d65aaa95e..766e88b2f5 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/register_mma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/register_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `register_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_arithmetic.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_arithmetic.rs
index a228c4b86d..43676af093 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_arithmetic.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_arithmetic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `scalar_arithmetic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_conversion.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_conversion.rs
index d0ad70c41c..45ccb2d478 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_conversion.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `scalar_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_math.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_math.rs
index 90726a69e7..ea1a765cb8 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_math.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/scalar_math.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `scalar_math` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sparse_mma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sparse_mma.rs
index 6cf699ea72..3e48ff5889 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sparse_mma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `sparse_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sreg.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sreg.rs
index 7ba1abc20c..4e2fa3ace6 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sreg.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `sreg` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/stmatrix.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/stmatrix.rs
index 8c2e34bd8f..77cf005293 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/stmatrix.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `stmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sync.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sync.rs
index 0e2925c790..7c67982816 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sync.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/sync.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `sync` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tcgen05.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tcgen05.rs
index 71a2fcd9bd..4dc9842f60 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tcgen05.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tcgen05.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `tcgen05` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tma.rs
index 023347d56c..2acc775bc3 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `tma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/vote.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/vote.rs
index bf76ce6cd9..2a3002383a 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/vote.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/vote.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `vote` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_barrier.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_barrier.rs
index 481095a476..98ad159499 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_barrier.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `warp_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_match.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_match.rs
index ed61420ea6..f33fee4ce7 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_match.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_match.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `warp_match` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_shuffle.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_shuffle.rs
index 6698b2e851..fe43a1c911 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_shuffle.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/warp_shuffle.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `warp_shuffle` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/wgmma_control.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/wgmma_control.rs
index f587d63acd..5da7982e82 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/wgmma_control.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tests/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated dispatch-path tests: `wgmma_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tma.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tma.rs
index 7d87b7e96a..8faf0ff889 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tma.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `tma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/vote.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/vote.rs
index 016a3ae61b..8fc9a547b7 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/vote.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/vote.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `vote` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_barrier.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_barrier.rs
index ea2efec42c..41199d8568 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_barrier.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `warp_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_match.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_match.rs
index 7da0eb23dd..2c0c6317cb 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_match.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_match.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `warp_match` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_shuffle.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_shuffle.rs
index f445adee0d..33310d9088 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_shuffle.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/warp_shuffle.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `warp_shuffle` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/wgmma_control.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/wgmma_control.rs
index 966b12094d..e0a203af54 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/wgmma_control.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/generated/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated raw/compatibility dispatch arms: `wgmma_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/integer_minmax.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/integer_minmax.rs
new file mode 100644
index 0000000000..13f3622085
--- /dev/null
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/integer_minmax.rs
@@ -0,0 +1,370 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+//! Primitive integer `Ord::min` and `Ord::max` intrinsics.
+
+use super::super::helpers;
+use crate::error::{TranslationErr, TranslationResult};
+use crate::translator::{rvalue, types, values::ValueMap};
+use dialect_mir::{
+    ops::{MirCallOp, MirLtOp},
+    rust_intrinsics,
+};
+use pliron::basic_block::BasicBlock;
+use pliron::builtin::{
+    attributes::StringAttr,
+    types::{IntegerType, Signedness},
+};
+use pliron::context::{Context, Ptr};
+use pliron::input_err;
+use pliron::location::{Located, Location};
+use pliron::op::Op;
+use pliron::operation::Operation;
+use pliron::r#type::{TypeHandle, Typed};
+use pliron::value::Value;
+use rustc_public::mir;
+
+/// Integer ordering intrinsics emitted by libcore.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum RustIntegerMinMaxIntrinsic {
+    /// Select the smaller operand.
+    Min,
+    /// Select the larger operand.
+    Max,
+}
+
+impl RustIntegerMinMaxIntrinsic {
+    /// Recognize exact libcore intrinsic paths, including the std reexports.
+    pub fn from_core_path(name: &str) -> Option<Self> {
+        match name {
+            "core::intrinsics::integer_min" | "std::intrinsics::integer_min" => Some(Self::Min),
+            "core::intrinsics::integer_max" | "std::intrinsics::integer_max" => Some(Self::Max),
+            _ => None,
+        }
+    }
+}
+
+/// Match core's specified fallback: `a < b` selects `(a, b)` for min and
+/// `(b, a)` for max. Keeping the original integer types lets existing compare
+/// lowering recover signedness, without narrowing any width (including 128).
+fn emit_integer_minmax(
+    ctx: &mut Context,
+    intrinsic: RustIntegerMinMaxIntrinsic,
+    args: &[Value],
+    return_type: TypeHandle,
+    block: Ptr<BasicBlock>,
+    prev_op: Option<Ptr<Operation>>,
+    loc: Location,
+) -> TranslationResult<Ptr<Operation>> {
+    let [lhs, rhs] = args else {
+        return input_err!(
+            loc,
+            TranslationErr::type_error("integer min/max expects two operands")
+        );
+    };
+    let valid_integer = return_type
+        .deref(ctx)
+        .downcast_ref::<IntegerType>()
+        .is_some_and(|ty| {
+            matches!(ty.width(), 8 | 16 | 32 | 64 | 128)
+                && matches!(ty.signedness(), Signedness::Signed | Signedness::Unsigned)
+        });
+    if !valid_integer || lhs.get_type(ctx) != return_type || rhs.get_type(ctx) != return_type {
+        return input_err!(
+            loc,
+            TranslationErr::type_error(
+                "integer min/max requires matching primitive integer operands and result"
+            )
+        );
+    }
+
+    let bool_type = types::get_bool_type(ctx).into();
+    let cmp = Operation::new(
+        ctx,
+        MirLtOp::get_concrete_op_info(),
+        vec![bool_type],
+        vec![*lhs, *rhs],
+        vec![],
+        0,
+    );
+    cmp.deref_mut(ctx).set_loc(loc.clone());
+    helpers::insert_op(ctx, cmp, block, prev_op);
+    let condition = cmp.deref(ctx).get_result(0);
+    let (if_true, if_false) = match intrinsic {
+        RustIntegerMinMaxIntrinsic::Min => (*lhs, *rhs),
+        RustIntegerMinMaxIntrinsic::Max => (*rhs, *lhs),
+    };
+    // This existing placeholder lowers to an ordinary LLVM select; it carries
+    // no unpredictable metadata or other optimization promises.
+    let select = Operation::new(
+        ctx,
+        MirCallOp::get_concrete_op_info(),
+        vec![return_type],
+        vec![condition, if_true, if_false],
+        vec![],
+        0,
+    );
+    select.deref_mut(ctx).set_loc(loc);
+    select.deref_mut(ctx).attributes.set(
+        "callee".try_into().unwrap(),
+        StringAttr::new(rust_intrinsics::CALLEE_SELECT_UNPREDICTABLE.into()),
+    );
+    select.insert_after(ctx, cmp);
+    Ok(select)
+}
+
+/// Translate integer min/max using the ordinary compare and select lowering.
+pub fn emit_rust_integer_minmax_intrinsic(
+    ctx: &mut Context,
+    body: &mir::Body,
+    intrinsic: RustIntegerMinMaxIntrinsic,
+    args: &[mir::Operand],
+    destination: &mir::Place,
+    target: &Option<usize>,
+    block_ptr: Ptr<BasicBlock>,
+    prev_op: Option<Ptr<Operation>>,
+    value_map: &mut ValueMap,
+    block_map: &[Ptr<BasicBlock>],
+    loc: Location,
+) -> TranslationResult<Ptr<Operation>> {
+    let return_type = types::translate_destination_type(ctx, body, destination, &loc)?;
+    let mut values = Vec::with_capacity(args.len());
+    let mut last_op = prev_op;
+    for arg in args {
+        let (value, last) =
+            rvalue::translate_operand(ctx, body, arg, value_map, block_ptr, last_op, loc.clone())?;
+        values.push(value);
+        last_op = last;
+    }
+    let (prepared, last_op) = helpers::prepare_destination_write(
+        ctx,
+        body,
+        destination,
+        value_map,
+        block_ptr,
+        last_op,
+        loc.clone(),
+    )?;
+    let select = emit_integer_minmax(
+        ctx,
+        intrinsic,
+        &values,
+        return_type,
+        block_ptr,
+        last_op,
+        loc.clone(),
+    )?;
+    let result = select.deref(ctx).get_result(0);
+    helpers::emit_prepared_result_and_goto(
+        ctx,
+        prepared,
+        result,
+        target,
+        block_ptr,
+        select,
+        value_map,
+        block_map,
+        loc,
+        "integer min/max call without target block",
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use llvm_export::{attributes::ICmpPredicateAttr, ops as llvm};
+    use pliron::builtin::{
+        attributes::TypeAttr, op_interfaces::SymbolOpInterface, ops::ModuleOp, types::FunctionType,
+    };
+    use pliron::common_traits::Verify;
+    use pliron::linked_list::ContainsLinkedList;
+
+    #[test]
+    fn recognizes_only_integer_intrinsic_paths() {
+        for prefix in ["core", "std"] {
+            assert_eq!(
+                RustIntegerMinMaxIntrinsic::from_core_path(&format!(
+                    "{prefix}::intrinsics::integer_min"
+                )),
+                Some(RustIntegerMinMaxIntrinsic::Min)
+            );
+            assert_eq!(
+                RustIntegerMinMaxIntrinsic::from_core_path(&format!(
+                    "{prefix}::intrinsics::integer_max"
+                )),
+                Some(RustIntegerMinMaxIntrinsic::Max)
+            );
+        }
+        for path in [
+            "user::intrinsics::integer_min",
+            "core::intrinsics::integer_minimum",
+            "core::intrinsics::integer_max_extra",
+            "core::cmp::min",
+            "core::intrinsics::minimum_number_nsz",
+        ] {
+            assert_eq!(RustIntegerMinMaxIntrinsic::from_core_path(path), None);
+        }
+    }
+
+    #[test]
+    fn lowers_minmax_with_original_width_signedness_and_selected_operands() {
+        for width in [8, 16, 32, 64, 128] {
+            for (signedness, predicate) in [
+                (Signedness::Signed, ICmpPredicateAttr::SLT),
+                (Signedness::Unsigned, ICmpPredicateAttr::ULT),
+            ] {
+                for (intrinsic, true_index, false_index) in [
+                    (RustIntegerMinMaxIntrinsic::Min, 0, 1),
+                    (RustIntegerMinMaxIntrinsic::Max, 1, 0),
+                ] {
+                    let mut ctx = Context::new();
+                    dialect_mir::register(&mut ctx);
+                    mir_lower::register(&mut ctx);
+                    let ty = IntegerType::get(&ctx, width, signedness).into();
+                    let module =
+                        ModuleOp::new(&mut ctx, "minmax".try_into().unwrap()).get_operation();
+                    let func_type = FunctionType::get(&ctx, vec![ty, ty], vec![ty]);
+                    let func = Operation::new(
+                        &mut ctx,
+                        dialect_mir::ops::MirFuncOp::get_concrete_op_info(),
+                        vec![],
+                        vec![],
+                        vec![],
+                        1,
+                    );
+                    let mir_func = dialect_mir::ops::MirFuncOp::new(
+                        &mut ctx,
+                        func,
+                        TypeAttr::new(func_type.into()),
+                    );
+                    mir_func.set_symbol_name(&mut ctx, "minmax".try_into().unwrap());
+                    let module_region = module.deref(&ctx).get_region(0);
+                    let module_block = module_region.deref(&ctx).iter(&ctx).next().unwrap();
+                    func.insert_at_back(module_block, &ctx);
+                    let region = func.deref(&ctx).get_region(0);
+                    let block = BasicBlock::new(&mut ctx, None, vec![ty, ty]);
+                    block.insert_at_back(region, &ctx);
+                    let args = [
+                        block.deref(&ctx).get_argument(0),
+                        block.deref(&ctx).get_argument(1),
+                    ];
+                    let select = emit_integer_minmax(
+                        &mut ctx,
+                        intrinsic,
+                        &args,
+                        ty,
+                        block,
+                        None,
+                        Location::Unknown,
+                    )
+                    .unwrap();
+                    let result = select.deref(&ctx).get_result(0);
+                    let ret = Operation::new(
+                        &mut ctx,
+                        dialect_mir::ops::MirReturnOp::get_concrete_op_info(),
+                        vec![],
+                        vec![result],
+                        vec![],
+                        0,
+                    );
+                    ret.insert_after(&ctx, select);
+                    mir_lower::lower_mir_to_llvm(&mut ctx, module).unwrap();
+                    module.deref(&ctx).verify(&ctx).unwrap();
+                    let func = module_block.deref(&ctx).iter(&ctx).next().unwrap();
+                    let region = func.deref(&ctx).get_region(0);
+                    let operations: Vec<_> = region
+                        .deref(&ctx)
+                        .iter(&ctx)
+                        .flat_map(|block| block.deref(&ctx).iter(&ctx))
+                        .collect();
+                    let compare = operations
+                        .iter()
+                        .find_map(|op| Operation::get_op::<llvm::ICmpOp>(*op, &ctx))
+                        .unwrap();
+                    assert_eq!(compare.predicate(&ctx), predicate);
+                    let compare = compare.get_operation();
+                    let select = operations
+                        .iter()
+                        .find_map(|op| Operation::get_op::<llvm::SelectOp>(*op, &ctx))
+                        .unwrap()
+                        .get_operation();
+                    assert_eq!(
+                        select.deref(&ctx).get_operand(0),
+                        compare.deref(&ctx).get_result(0)
+                    );
+                    assert_eq!(
+                        select.deref(&ctx).get_operand(1),
+                        block.deref(&ctx).get_argument(true_index)
+                    );
+                    assert_eq!(
+                        select.deref(&ctx).get_operand(2),
+                        block.deref(&ctx).get_argument(false_index)
+                    );
+                    assert_eq!(
+                        compare.deref(&ctx).get_operand(0),
+                        block.deref(&ctx).get_argument(0)
+                    );
+                    assert_eq!(
+                        compare.deref(&ctx).get_operand(1),
+                        block.deref(&ctx).get_argument(1)
+                    );
+                    assert_eq!(
+                        select
+                            .deref(&ctx)
+                            .get_result(0)
+                            .get_type(&ctx)
+                            .deref(&ctx)
+                            .downcast_ref::<IntegerType>()
+                            .unwrap()
+                            .width(),
+                        width
+                    );
+                }
+            }
+        }
+    }
+
+    #[test]
+    fn rejects_invalid_integer_minmax_signatures_before_emission() {
+        let mut ctx = Context::new();
+        dialect_mir::register(&mut ctx);
+        let integer: TypeHandle = IntegerType::get(&ctx, 32, Signedness::Unsigned).into();
+        let signed: TypeHandle = IntegerType::get(&ctx, 32, Signedness::Signed).into();
+        let wide: TypeHandle = IntegerType::get(&ctx, 64, Signedness::Unsigned).into();
+        let boolean: TypeHandle = IntegerType::get(&ctx, 1, Signedness::Signless).into();
+        let signless: TypeHandle = IntegerType::get(&ctx, 32, Signedness::Signless).into();
+        let odd_width: TypeHandle = IntegerType::get(&ctx, 24, Signedness::Unsigned).into();
+        let float: TypeHandle = pliron::builtin::types::FP32Type::get(&ctx).into();
+        for (args, result) in [
+            (vec![integer], integer),
+            (vec![integer, integer, integer], integer),
+            (vec![integer, wide], integer),
+            (vec![integer, signed], integer),
+            (vec![integer, integer], wide),
+            (vec![boolean, boolean], boolean),
+            (vec![signless, signless], signless),
+            (vec![odd_width, odd_width], odd_width),
+            (vec![float, float], float),
+        ] {
+            let block = BasicBlock::new(&mut ctx, None, args.clone());
+            let values: Vec<_> = (0..args.len())
+                .map(|index| block.deref(&ctx).get_argument(index))
+                .collect();
+            assert!(
+                emit_integer_minmax(
+                    &mut ctx,
+                    RustIntegerMinMaxIntrinsic::Min,
+                    &values,
+                    result,
+                    block,
+                    None,
+                    Location::Unknown
+                )
+                .is_err()
+            );
+            assert!(block.deref(&ctx).iter(&ctx).next().is_none());
+        }
+    }
+}
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/mod.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/mod.rs
index f70f591f38..5c50b07cde 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/mod.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/intrinsics/mod.rs
@@ -12,6 +12,7 @@
 //! |--------------|------------------------------------------------------------------------------|
 //! | `generated`  | Every catalog intrinsic, dispatched by canonical and compatibility path      |
 //! | `indexing`   | `threadIdx_*`, `blockIdx_*`, `index_1d`, `index_2d::<S>`, `index_2d_runtime` |
+//! | `integer_minmax` | Rust compiler integer minimum and maximum intrinsics                    |
 //! | `memory`     | `SharedArray`, `stmatrix_*`, type conversions                                |
 //! | `atomic`     | Atomic read-modify-write and compare-exchange                                |
 //! | `wgmma`      | Hopper WGMMA matrix operations                                               |
@@ -51,6 +52,7 @@ pub mod float_math;
 pub mod generated;
 pub mod iket;
 pub mod indexing;
+pub mod integer_minmax;
 pub mod layout;
 pub mod memory;
 pub mod saturating;
diff --git a/cuda-oxide/crates/mir-importer/src/translator/terminator/mod.rs b/cuda-oxide/crates/mir-importer/src/translator/terminator/mod.rs
index 34f8d9889c..15321134d3 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/terminator/mod.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/terminator/mod.rs
@@ -2761,6 +2761,26 @@ fn try_dispatch_intrinsic(
         )?));
     }
 
+    if let Some(intrinsic) =
+        intrinsics::integer_minmax::RustIntegerMinMaxIntrinsic::from_core_path(name)
+    {
+        return Ok(Some(
+            intrinsics::integer_minmax::emit_rust_integer_minmax_intrinsic(
+                ctx,
+                body,
+                intrinsic,
+                args,
+                destination,
+                target,
+                block_ptr,
+                prev_op,
+                value_map,
+                block_map,
+                loc,
+            )?,
+        ));
+    }
+
     if let Some(intrinsic) = intrinsics::saturating::RustSaturatingIntrinsic::from_core_path(name) {
         return Ok(Some(
             intrinsics::saturating::emit_rust_saturating_intrinsic(
@@ -2827,7 +2847,13 @@ fn try_dispatch_intrinsic(
         )?));
     }
 
-    if let Some(intrinsic) = intrinsics::float_math::RustFloatMathIntrinsic::from_core_path(name) {
+    if let Some(intrinsic) =
+        intrinsics::float_math::RustFloatMathIntrinsic::from_core_path_with_substs(
+            name,
+            type_substs,
+        )
+        .map_err(|error| input_error!(loc.clone(), TranslationErr::unsupported(error)))?
+    {
         return Ok(Some(
             intrinsics::float_math::emit_rust_float_math_intrinsic(
                 ctx,
diff --git a/cuda-oxide/crates/mir-importer/src/translator/types.rs b/cuda-oxide/crates/mir-importer/src/translator/types.rs
index ee5c23b8e7..0b9c79aaca 100644
--- a/cuda-oxide/crates/mir-importer/src/translator/types.rs
+++ b/cuda-oxide/crates/mir-importer/src/translator/types.rs
@@ -859,8 +859,10 @@ pub fn translate_type(
                             // Total struct size (bytes)
                             let size: u64 = shape.size.bytes() as u64;
                             let abi_kind = if adt_def.repr().flags.is_transparent
-                                && matches!(&shape.abi, rustc_public::abi::ValueAbi::Scalar(_))
-                            {
+                                && matches!(
+                                    &shape.value_repr,
+                                    rustc_public::abi::ValueRepr::Scalar(_)
+                                ) {
                                 StructAbiKind::TransparentScalar
                             } else {
                                 StructAbiKind::Aggregate
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/active_mask.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/active_mask.rs
index 95b3f0daf5..8406c947cb 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/active_mask.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/active_mask.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `active_mask` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/clc.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/clc.rs
index fc7b13775f..7234063462 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/clc.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/clc.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `clc` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_barrier.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_barrier.rs
index 79e59a186a..028bc9bbe3 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_barrier.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `cluster_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_memory.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_memory.rs
index 809bdd2fe6..3e1aa64684 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_memory.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cluster_memory.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `cluster_memory` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cp_async.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cp_async.rs
index 30ec31ca6c..c57daa967f 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cp_async.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/cp_async.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `cp_async` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/debug_control.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/debug_control.rs
index cc37b0bcac..e59ef90367 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/debug_control.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/debug_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `debug_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/dotprod.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/dotprod.rs
index 5f1c71aa9e..10bb6fe166 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/dotprod.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/dotprod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `dotprod` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/elect.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/elect.rs
index 49d0485ba0..e4d3bd35ca 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/elect.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/elect.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `elect` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/execution_control.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/execution_control.rs
index eb33f917ed..89dbae811c 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/execution_control.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/execution_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `execution_control` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/extended_minmax.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/extended_minmax.rs
index 64976d1c34..60e8ee934c 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/extended_minmax.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/extended_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `extended_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/integer_minmax.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/integer_minmax.rs
index dbfe47cc8c..435dd821d3 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/integer_minmax.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/integer_minmax.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `integer_minmax` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/ldmatrix.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/ldmatrix.rs
index 3c029721fe..d12cc4e355 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/ldmatrix.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/ldmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `ldmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_basic.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_basic.rs
index 760b526608..3555905110 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_basic.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_basic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `mbarrier_basic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_extended.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_extended.rs
index 6a2e00d2da..75541afe5d 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_extended.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mbarrier_extended.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `mbarrier_extended` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mod.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mod.rs
index fdaa4bf21c..3215b3beee 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mod.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/mod.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces for admitted CUDA intrinsic families.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/movmatrix.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/movmatrix.rs
index ce5d980b17..22f41b615a 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/movmatrix.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/movmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `movmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_alu.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_alu.rs
index 3d9d39e5ca..1c15e63abb 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_alu.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_alu.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `packed_alu` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_atomic.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_atomic.rs
index 8b10729fce..37a5f5529f 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_atomic.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_atomic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `packed_atomic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_conversion.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_conversion.rs
index 7f9e1b49fb..cd88761a89 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_conversion.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/packed_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `packed_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/prmt.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/prmt.rs
index 7fcacbd0b9..d45ea8b4da 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/prmt.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/prmt.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `prmt` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/redux.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/redux.rs
index a776f60f9d..ba3730d318 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/redux.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/redux.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `redux` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/register_mma.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/register_mma.rs
index 3258cf9bd4..8e978ea5d6 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/register_mma.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/register_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `register_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_arithmetic.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_arithmetic.rs
index eb2f26f21f..25a0be1634 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_arithmetic.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_arithmetic.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `scalar_arithmetic` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_conversion.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_conversion.rs
index 0d5779dab4..b44db8874a 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_conversion.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_conversion.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `scalar_conversion` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_math.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_math.rs
index 329b2242f7..ce0b651199 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_math.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/scalar_math.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `scalar_math` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sparse_mma.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sparse_mma.rs
index 6d6b104715..38d6078474 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sparse_mma.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sparse_mma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `sparse_mma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sreg.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sreg.rs
index e8c341fbba..441e699695 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sreg.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sreg.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `sreg` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/stmatrix.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/stmatrix.rs
index 639c29ce43..8e465421a3 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/stmatrix.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/stmatrix.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `stmatrix` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sync.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sync.rs
index 0385565067..05fd89a47e 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sync.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/sync.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `sync` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tcgen05.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tcgen05.rs
index 6819daab9d..8863051c4e 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tcgen05.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tcgen05.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `tcgen05` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tma.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tma.rs
index 5ba6cbc944..3ba049ba8f 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tma.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/tma.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `tma` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/vote.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/vote.rs
index f385fe8109..9de55faf9b 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/vote.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/vote.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `vote` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_barrier.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_barrier.rs
index 6d49318480..7e15564c79 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_barrier.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_barrier.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `warp_barrier` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_match.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_match.rs
index fd4aa60aa9..b29139f764 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_match.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_match.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `warp_match` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_shuffle.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_shuffle.rs
index d6822754f5..b84d82d837 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_shuffle.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/warp_shuffle.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `warp_shuffle` intrinsics.
 
diff --git a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/wgmma_control.rs b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/wgmma_control.rs
index c60e551e5a..81ddb79866 100644
--- a/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/wgmma_control.rs
+++ b/cuda-oxide/crates/mir-lower/src/convert/generated_intrinsics/wgmma_control.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated conversion interfaces: `wgmma_control` intrinsics.
 
diff --git a/cuda-oxide/crates/rustc-codegen-cuda/README.md b/cuda-oxide/crates/rustc-codegen-cuda/README.md
index c1a7f2d550..372dc3d8a8 100644
--- a/cuda-oxide/crates/rustc-codegen-cuda/README.md
+++ b/cuda-oxide/crates/rustc-codegen-cuda/README.md
@@ -194,4 +194,4 @@ cargo oxide pipeline <example_name>
 
 ### Nightly Rust
 
-The backend uses `#![feature(rustc_private)]` and pins to a specific nightly via `rust-toolchain.toml`. The workspace currently uses `nightly-2026-08-28`.
+The backend uses `#![feature(rustc_private)]` and pins to a specific nightly via `rust-toolchain.toml`. The workspace currently uses `nightly-2026-10-11`.
diff --git a/cuda-oxide/crates/rustc-codegen-cuda/examples/atomics/README.md b/cuda-oxide/crates/rustc-codegen-cuda/examples/atomics/README.md
index 2d839810ca..a99a7bdca3 100644
--- a/cuda-oxide/crates/rustc-codegen-cuda/examples/atomics/README.md
+++ b/cuda-oxide/crates/rustc-codegen-cuda/examples/atomics/README.md
@@ -14,7 +14,7 @@ Atomic operations require **LLVM 22 or newer** for correct syncscope
 generation. Without it, scopes (`.gpu`, `.cta`, `.sys`) will be missing
 from generated PTX.
 
-The pinned Rust toolchain (`nightly-2026-08-28`) ships LLVM 23 with NVPTX
+The pinned Rust toolchain (`nightly-2026-10-11`) ships LLVM 23 with NVPTX
 enabled via the `llvm-tools` component, so the default onboarding path
 already satisfies this requirement:
 
@@ -83,7 +83,7 @@ CUDA_OXIDE_VERBOSE=1 cargo oxide run atomics
 
 ### Floating-point add lowering
 
-With the pinned `nightly-2026-08-28` toolchain's LLVM 23 NVPTX defaults,
+With the pinned `nightly-2026-10-11` toolchain's LLVM 23 NVPTX defaults,
 `atomic_f32_fetch_add_test` uses a compare-and-swap (CAS) loop for global memory.
 Generic-address-space `f32` adds also use a loop; shared `f32` and `f64` adds
 still use native instructions on supported targets. For example:
diff --git a/cuda-oxide/crates/rustc-codegen-cuda/src/collector.rs b/cuda-oxide/crates/rustc-codegen-cuda/src/collector.rs
index c361689734..0b2bb713ab 100644
--- a/cuda-oxide/crates/rustc-codegen-cuda/src/collector.rs
+++ b/cuda-oxide/crates/rustc-codegen-cuda/src/collector.rs
@@ -2272,7 +2272,7 @@ impl<'tcx> DeviceCollector<'tcx> {
             .with_help(
                 "use the cuda-intrinsics crate generated by this cuda-oxide revision, then rebuild the device crate",
             )
-            .emit()
+            .emit_fatal()
     }
 
     /// Emits the heap-allocation diagnostic (issue #108) and aborts.
@@ -2308,7 +2308,7 @@ impl<'tcx> DeviceCollector<'tcx> {
                 "store the data in a fixed-size array `[T; N]` instead, or in a \
                  `SharedArray` when the scratch space should be shared by the thread block",
             )
-            .emit()
+            .emit_fatal()
     }
 
     /// Refuses a device-reachable function that only makes sense on the host
diff --git a/cuda-oxide/crates/rustc-codegen-cuda/src/device_codegen.rs b/cuda-oxide/crates/rustc-codegen-cuda/src/device_codegen.rs
index 7e62630445..457b7c09da 100644
--- a/cuda-oxide/crates/rustc-codegen-cuda/src/device_codegen.rs
+++ b/cuda-oxide/crates/rustc-codegen-cuda/src/device_codegen.rs
@@ -103,6 +103,7 @@ use rustc_middle::mir::interpret::{AllocId, GlobalAlloc, Scalar};
 use rustc_middle::mir::visit::Visitor;
 use rustc_middle::mir::{ConstOperand, ConstValue, Location};
 use rustc_middle::mono::MonoItem;
+use rustc_middle::ty::consts::ConstExt;
 use rustc_middle::ty::layout::{LayoutCx, LayoutOf};
 use rustc_middle::ty::{EarlyBinder, Instance, InstanceKind, ShimKind, TypingEnv};
 use rustc_middle::ty::{Ty, TyCtxt, TyKind};
@@ -217,7 +218,7 @@ fn rustc_ty_to_device_extern_type<'tcx>(
                     "failed to compute layout for repr(transparent) device-extern type `{ty}`: {err:?}"
                 )
             })?;
-            let peeled = layout.peel_transparent_wrappers(&layout_cx);
+            let peeled = layout.peel_transparent_wrappers_from_non_1zst(&layout_cx);
 
             // A transparent type with no peelable non-1ZST field is not a
             // scalar/pointer wrapper that this device-extern ABI can represent.
@@ -1070,10 +1071,7 @@ pub fn generate_device_code<'tcx>(
         .iter()
         .map(|func| {
             let def_id = func.instance.def_id();
-            matches!(
-                tcx.codegen_fn_attrs(def_id).inline,
-                rustc_hir::attrs::InlineAttr::Always | rustc_hir::attrs::InlineAttr::Force { .. }
-            )
+            tcx.codegen_fn_attrs(def_id).inline.always()
         })
         .collect();
     let device_mono_reachability: Vec<crate::collector::DeviceMonoReachability> = functions
diff --git a/cuda-oxide/crates/rustc-codegen-cuda/src/generated_intrinsics.rs b/cuda-oxide/crates/rustc-codegen-cuda/src/generated_intrinsics.rs
index 2c0b79d2e7..d0f9932736 100644
--- a/cuda-oxide/crates/rustc-codegen-cuda/src/generated_intrinsics.rs
+++ b/cuda-oxide/crates/rustc-codegen-cuda/src/generated_intrinsics.rs
@@ -4,8 +4,8 @@
  */
 
 // @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-// LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+// Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+// LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 //! Generated collector predicates for intrinsic placeholder functions.
 
diff --git a/cuda-oxide/crates/rustc-codegen-cuda/src/lib.rs b/cuda-oxide/crates/rustc-codegen-cuda/src/lib.rs
index 8deaf0cf0f..e0e7366849 100644
--- a/cuda-oxide/crates/rustc-codegen-cuda/src/lib.rs
+++ b/cuda-oxide/crates/rustc-codegen-cuda/src/lib.rs
@@ -337,7 +337,7 @@ use rustc_middle::dep_graph::WorkProductMap;
 use rustc_middle::ty::TyCtxt;
 use rustc_middle::ty::print::with_no_trimmed_paths;
 use rustc_session::config::OutputFilenames;
-use rustc_session::{IncrCompSession, Session};
+use rustc_session::{CodegenBackendInit, EarlySession, IncrCompSession, Session};
 use std::any::Any;
 use std::collections::BTreeSet;
 use std::path::{Path, PathBuf};
@@ -461,12 +461,12 @@ impl CodegenBackend for CudaCodegenBackend {
         "cuda"
     }
 
-    fn init(&self, sess: &Session) {
+    fn init(&mut self, sess: &EarlySession) -> CodegenBackendInit {
         // Note: Don't log here - init() is called for ALL crates including dependencies.
         // We log in codegen_crate() only when there are kernels to compile.
 
         // Initialize the underlying LLVM backend
-        self.llvm_backend.init(sess);
+        self.llvm_backend.init(sess)
     }
 
     fn print_version(&self) {
@@ -481,7 +481,7 @@ impl CodegenBackend for CudaCodegenBackend {
         self.llvm_backend.target_cpu(sess)
     }
 
-    fn target_config(&self, sess: &Session) -> rustc_codegen_ssa::TargetConfig {
+    fn target_config(&self, sess: &EarlySession) -> rustc_codegen_ssa::TargetConfig {
         self.llvm_backend.target_config(sess)
     }
 
@@ -831,7 +831,6 @@ impl CodegenBackend for CudaCodegenBackend {
                 assembly: None,
                 llvm_ir: None,
                 global_asm_object: None,
-                links_from_incr_cache: Vec::new(),
             });
         }
         (compiled_modules, work_products)
diff --git a/cuda-oxide/cuda-oxide-book/appendix/building-from-source.md b/cuda-oxide/cuda-oxide-book/appendix/building-from-source.md
index 598843eb77..1167c28016 100644
--- a/cuda-oxide/cuda-oxide-book/appendix/building-from-source.md
+++ b/cuda-oxide/cuda-oxide-book/appendix/building-from-source.md
@@ -10,7 +10,7 @@ from a fresh checkout. If you just want to run an example, the
 
 | Dependency       | Version                       | Purpose                                                     |
 |:-----------------|:----------------------------- |:------------------------------------------------------------|
-| **Rust nightly** | `nightly-2026-08-28` (pinned) | Compiler toolchain with `rustc-dev` for the codegen backend |
+| **Rust nightly** | `nightly-2026-10-11` (pinned) | Compiler toolchain with `rustc-dev` for the codegen backend |
 | **CUDA Toolkit** | 13.0+ (with cuRAND headers)   | Driver API, `nvcc`, PTX assembler; `curand.h` for bindgen   |
 | **Clang**        | 21+ (`clang-21` pkg)          | `bindgen` in host `cuda-bindings` needs clang's headers     |
 | **Linux**        | Tested on Ubuntu 24.04        | Windows and macOS are not supported                         |
@@ -33,15 +33,15 @@ stable:
 ```toml
 # cuda-oxide/rust-toolchain.toml
 [toolchain]
-channel = "nightly-2026-08-28"
+channel = "nightly-2026-10-11"
 components = ["rust-src", "rustc-dev", "rust-analyzer", "clippy", "rustfmt", "llvm-tools"]
 ```
 
 If you need to install manually:
 
 ```bash
-rustup toolchain install nightly-2026-08-28
-rustup component add rust-src rustc-dev rust-analyzer clippy rustfmt llvm-tools --toolchain nightly-2026-08-28
+rustup toolchain install nightly-2026-10-11
+rustup component add rust-src rustc-dev rust-analyzer clippy rustfmt llvm-tools --toolchain nightly-2026-10-11
 ```
 
 `rust-src` provides the standard library source for cross-compilation,
@@ -66,7 +66,7 @@ required for `ptxas` and header files, but you will not be able to run kernels.
 ## Install LLVM (usually optional)
 
 The codegen pipeline emits LLVM IR and invokes `llc` to produce PTX. The
-pinned Rust toolchain (`nightly-2026-08-28`) already ships LLVM 23 with the
+pinned Rust toolchain (`nightly-2026-10-11`) already ships LLVM 23 with the
 NVPTX backend enabled via the `llvm-tools` component, so the recommended
 path is:
 
@@ -155,7 +155,7 @@ build process. `cargo-oxide` handles building it transparently.
 pipeline. Inside the repo, it works via a workspace alias. For standalone use, install it with the pinned nightly toolchain:
 
 ```bash
-cargo +nightly-2026-08-28 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
+cargo +nightly-2026-10-11 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
 ```
 
 On first run, `cargo-oxide` automatically fetches and builds the codegen backend
@@ -277,7 +277,7 @@ repository/
 
 `error[E0463]: can't find crate for rustc_middle`
 : You are missing the `rustc-dev` component. Run:
-  `rustup component add rustc-dev --toolchain nightly-2026-08-28`.
+  `rustup component add rustc-dev --toolchain nightly-2026-10-11`.
 
 CUDA driver version mismatch
 : The toolkit version (compile-time) and driver version (runtime) must be
diff --git a/cuda-oxide/cuda-oxide-book/compiler/lowering-pipeline.md b/cuda-oxide/cuda-oxide-book/compiler/lowering-pipeline.md
index 93217b0277..eb0cc7e4a4 100644
--- a/cuda-oxide/cuda-oxide-book/compiler/lowering-pipeline.md
+++ b/cuda-oxide/cuda-oxide-book/compiler/lowering-pipeline.md
@@ -572,7 +572,7 @@ that cuda-oxide emits.
 | 5th      | `llc-21` on `PATH`                                        | Distro / `apt.llvm.org` install of LLVM 21.                           |
 | 6th      | `llc` on `PATH`                                           | Reporting fallback only; rejected at runtime if older than LLVM 21.   |
 
-The pinned Rust toolchain (`nightly-2026-08-28`) ships LLVM 23 with NVPTX
+The pinned Rust toolchain (`nightly-2026-10-11`) ships LLVM 23 with NVPTX
 enabled, so `rustup component add llvm-tools` is the recommended onboarding
 path. The PATH probes for `llc-23` / `llc-22` / `llc-21` are kept as a fallback for
 users with an existing LLVM install. If none of the probes succeed the
diff --git a/cuda-oxide/cuda-oxide-book/compiler/mlir-dialects.md b/cuda-oxide/cuda-oxide-book/compiler/mlir-dialects.md
index c3057e5280..8cff3c042a 100644
--- a/cuda-oxide/cuda-oxide-book/compiler/mlir-dialects.md
+++ b/cuda-oxide/cuda-oxide-book/compiler/mlir-dialects.md
@@ -303,11 +303,11 @@ they become `call` instructions to `@llvm.nvvm.*` intrinsics.
 
 ### Architecture Coverage
 
-At catalog SHA-256 `016c9d9b` (the stamp in every `ops/generated/` file
+At catalog SHA-256 `c243e568` (the stamp in every `ops/generated/` file
 header), the dialect holds 576 operations across 42 modules, and they come
 from two different places. The split is the first thing to know about it,
 because it decides where -- and whether -- you would add one. If the header
-stamp no longer starts with `016c9d9b`, the counts on this page predate the
+stamp no longer starts with `c243e568`, the counts on this page predate the
 catalog you are reading.
 
 **Hand-written**, directly under `cuda-oxide/crates/dialect-nvvm/src/ops/`. These are the
diff --git a/cuda-oxide/cuda-oxide-book/compiler/rustc-codegen-cuda.md b/cuda-oxide/cuda-oxide-book/compiler/rustc-codegen-cuda.md
index b9513030e5..683605591e 100644
--- a/cuda-oxide/cuda-oxide-book/compiler/rustc-codegen-cuda.md
+++ b/cuda-oxide/cuda-oxide-book/compiler/rustc-codegen-cuda.md
@@ -215,7 +215,7 @@ takes the `#[kernel]` macro's base name and, for generic and closure-
 generic instantiations, appends `_TID_<hex32>` where `<hex32>` is
 rustc's 128-bit type-id hash of the concrete generated kernel function item.
 That `FnDef` contains ordered type and const arguments. The host launcher
-computes the same hash through the `core::intrinsics::type_id` intrinsic
+computes the same hash through the `core::intrinsics::reflection::type_id` intrinsic
 (wrapped by `cuda_host::type_id_u128_of_val`), so both sides agree
 byte-for-byte within one unified build. Non-generic kernels keep their bare
 base name.
diff --git a/cuda-oxide/cuda-oxide-book/compiler/rustc-public.md b/cuda-oxide/cuda-oxide-book/compiler/rustc-public.md
index 737cb0fdb5..b157eda3d1 100644
--- a/cuda-oxide/cuda-oxide-book/compiler/rustc-public.md
+++ b/cuda-oxide/cuda-oxide-book/compiler/rustc-public.md
@@ -288,7 +288,7 @@ cuda-oxide pins to an exact nightly release via `rust-toolchain.toml`:
 
 ```toml
 [toolchain]
-channel = "nightly-2026-08-28"
+channel = "nightly-2026-10-11"
 components = ["rust-src", "rustc-dev", "rust-analyzer", "clippy", "rustfmt", "llvm-tools"]
 ```
 
diff --git a/cuda-oxide/cuda-oxide-book/getting-started/hello-gpu.md b/cuda-oxide/cuda-oxide-book/getting-started/hello-gpu.md
index 6c0325b9bc..6938672580 100644
--- a/cuda-oxide/cuda-oxide-book/getting-started/hello-gpu.md
+++ b/cuda-oxide/cuda-oxide-book/getting-started/hello-gpu.md
@@ -9,7 +9,7 @@ This section walks through installing cuda-oxide, creating a project, writing a
 If you haven't already, install the build tool with the pinned nightly toolchain:
 
 ```bash
-cargo +nightly-2026-08-28 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
+cargo +nightly-2026-10-11 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
 ```
 
 Verify that your environment is set up correctly:
diff --git a/cuda-oxide/cuda-oxide-book/getting-started/installation.md b/cuda-oxide/cuda-oxide-book/getting-started/installation.md
index 59ae7b28ae..8d6fa79faf 100644
--- a/cuda-oxide/cuda-oxide-book/getting-started/installation.md
+++ b/cuda-oxide/cuda-oxide-book/getting-started/installation.md
@@ -262,8 +262,8 @@ automatically. Commands at the repository root and in `cutile-rs/` use stable.
 If you need to install it manually:
 
 ```bash
-rustup toolchain install nightly-2026-08-28
-rustup component add rust-src rustc-dev rust-analyzer clippy rustfmt llvm-tools --toolchain nightly-2026-08-28
+rustup toolchain install nightly-2026-10-11
+rustup component add rust-src rustc-dev rust-analyzer clippy rustfmt llvm-tools --toolchain nightly-2026-10-11
 ```
 
 Three of these components are what the codegen backend and doctor need:
@@ -285,7 +285,7 @@ The other three are not needed to build: `rust-analyzer` powers IDE support, `cl
 **For use outside the repo** (your own projects), install it with the pinned nightly toolchain:
 
 ```bash
-cargo +nightly-2026-08-28 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
+cargo +nightly-2026-10-11 install --git https://github.com/NVlabs/cuda-oxide.git cargo-oxide
 ```
 
 On first run, `cargo-oxide` will automatically fetch and build the codegen backend. Subsequent runs reuse the cached build.
@@ -396,8 +396,8 @@ If everything is configured correctly, this compiles a Rust kernel to PTX, launc
 :::{tip}
 **Common issues:**
 
-- `No working llc found` -- prefer `rustup component add llvm-tools --toolchain nightly-2026-08-28`, or install LLVM 21+ (`sudo apt install llvm-21`), add `/usr/lib/llvm-21/bin` to your `PATH`, or set `CUDA_OXIDE_LLC=/usr/bin/llc-21`.
+- `No working llc found` -- prefer `rustup component add llvm-tools --toolchain nightly-2026-10-11`, or install LLVM 21+ (`sudo apt install llvm-21`), add `/usr/lib/llvm-21/bin` to your `PATH`, or set `CUDA_OXIDE_LLC=/usr/bin/llc-21`.
 - `'stddef.h' file not found` when building host `cuda-bindings` -- install clang dev headers: `sudo apt install clang-21` (or `libclang-common-21-dev`).
 - `cuda.h not found` -- Set `CUDA_TOOLKIT_PATH` to your CUDA install root, or ensure `/usr/local/cuda/include/cuda.h` exists.
-- `rust-src` / `llvm-tools` component missing -- Run `rustup component add rust-src llvm-tools --toolchain nightly-2026-08-28`.
+- `rust-src` / `llvm-tools` component missing -- Run `rustup component add rust-src llvm-tools --toolchain nightly-2026-10-11`.
 :::
diff --git a/cuda-oxide/flake.lock b/cuda-oxide/flake.lock
index 0f144f8898..d938292504 100644
--- a/cuda-oxide/flake.lock
+++ b/cuda-oxide/flake.lock
@@ -64,11 +64,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1787993548,
-        "narHash": "sha256-+IAEnmmx5YIhUWo0lp15jLLHchnXo5yKgWsi6C6Cf+0=",
+        "lastModified": 1791707771,
+        "narHash": "sha256-w3m8HtAVCfGTHw+gcCTAy7jFHSdGbokm8NngbeFrkek=",
         "owner": "oxalica",
         "repo": "rust-overlay",
-        "rev": "996e9b0b019a4a9eb9e9a5641aefa06d801b5895",
+        "rev": "c1cfa7a2c809d364ac1a7e9b3ec418afa3f19793",
         "type": "github"
       },
       "original": {
diff --git a/cuda-oxide/intrinsics/catalog.json b/cuda-oxide/intrinsics/catalog.json
index 1066e57eaf..9c57efa92b 100644
--- a/cuda-oxide/intrinsics/catalog.json
+++ b/cuda-oxide/intrinsics/catalog.json
@@ -5,13 +5,13 @@
   "generator_version": "0.1.0",
   "source": {
     "llvm_repository": "https://github.com/rust-lang/llvm-project.git",
-    "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-    "llvm_tblgen_version": "LLVM version 23.1.0",
-    "llvm_tblgen_source_revision": "16696adcd119e6ba9cc175207d984d7021211acb"
+    "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+    "llvm_tblgen_version": "LLVM version 23.1.3",
+    "llvm_tblgen_source_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019"
   },
   "inputs": {
-    "imported_sha256": "3f54245b231a5b6e19f3225710c3865b7e1a106242cce873250ef5fdc32e532e",
-    "overlay_sha256": "ecc7d0d7d70db737303abb6271666c67e3757a2727ba2191946a6d4e66e9f717",
+    "imported_sha256": "54d47b8e281dd953dc95fe16f834ff38164d9809171c9a1af097231b0bc45cab",
+    "overlay_sha256": "08f7a25510022bf48902cb4889b9f711a1cff31d1c5e8ce21c669d61afbb0310",
     "abi_ledger_sha256": "fc378e74ef6da50df31de62c3fafcb3264dfb414034e5852147359fcad4184eb",
     "evidence_sha256": [
       "cuda-12.0-libnvvm-12.0.140-ex2-f16.json:7102ea7d567ff2d2eb75a0384148d2e5eb6e605c26a5b98c71371fa3c46adef6",
@@ -73,65 +73,65 @@
       "cuda-13.3-libnvvm-13.3.73-tma-descriptor-control.json:ac44f485122404f9a900cfb94bff658160a6c23d9e2f8ef74bb751f044b848fc",
       "cuda-13.3-libnvvm-13.3.73-tma-reduce.json:f5df5bb51068966594660cb01b1b28c3b72cb004c456fcd8dff377ff0bee4fc0",
       "cuda-13.3-libnvvm-13.3.73-tma.json:8607f93605f215858a98075a0ea1394f2718d090538bcaf0a795e11b4cb66d47",
-      "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix.json:16b9e82da0f44ba327caf383ec107e45e2103bb3c8c5b66e412779b575a2c741",
-      "rust-llvm-23.1.0-16696adc-clc.json:afc13ecb0e9a316d5102e98beea67e921a89c2474e46dc0d8d1a9cf4a3b12d4c",
-      "rust-llvm-23.1.0-16696adc-cluster-barrier.json:2e3886e63765415c1f4467670adf9253a4733d0b292f13536f00461c1fd84f14",
-      "rust-llvm-23.1.0-16696adc-cluster-memory.json:13067b0e40a71e50e0a9b8e7b6b41d72f069bd95d97e6cf67906a8a7ac1d80af",
-      "rust-llvm-23.1.0-16696adc-cluster-sreg.json:f637df8caa13340748e1ed32c63c24c186c96f1eacb72e95a80aabb64dc4fade",
-      "rust-llvm-23.1.0-16696adc-cp-async-mbarrier.json:49bcc68b25dafe33941e53bfd9f8f7838d9ba9cd678c2b8349469c1ae2aa66d0",
-      "rust-llvm-23.1.0-16696adc-cp-async.json:191a81aacb8e52de15aced1d3ecd15269a14e9f15c351134357a71fe979a8ed0",
-      "rust-llvm-23.1.0-16696adc-debug-control.json:d73c53717cd7284ad64ec49e2b5bd0514f4bb9088cc17c1fe3cb566809d2030a",
-      "rust-llvm-23.1.0-16696adc-elect.json:8c8f7fdc3d6e85b89ec072d349bbb89e33632b5d3cf15a63456c6d2b74a770c0",
-      "rust-llvm-23.1.0-16696adc-execution-control.json:bd7043e20087c94e5fa90ebf32f634fe46958f5a33f3ae0e5167f54a8dc63729",
-      "rust-llvm-23.1.0-16696adc-extended-minmax.json:9ab3c99ff30d06ce44af46be5de9e5d960113b106bf53e90f4fd317763380c18",
-      "rust-llvm-23.1.0-16696adc-f32x2.json:2b8ac4914be67e2d97c730914713cad930628cf3ffef86e6114612b408d030e6",
-      "rust-llvm-23.1.0-16696adc-integer-minmax.json:bf56449724c0ed5325586222a6e03bc2f53cad25ba50ac894fd6e75e86296e48",
-      "rust-llvm-23.1.0-16696adc-mbarrier-basic.json:2663376dd611ba3c38bfdd00948ec45c6a4ad004e972f9392097eec7d19c8292",
-      "rust-llvm-23.1.0-16696adc-mbarrier-extended.json:c53bd6fcbf47833fc102e7a6c94ca53c858ffc7e09429841bf4385b42526eb02",
-      "rust-llvm-23.1.0-16696adc-packed-conversion-fp8-f16x2.json:ff5c15aa45ad7177f83a669e293faaf44cf7302d5fc2b0f4163245652933d9c1",
-      "rust-llvm-23.1.0-16696adc-packed-conversion-fp8.json:5d14d86b3ee6e17bcf6b07647c686649e97a27cc06dd45d8c31307b4705c4564",
-      "rust-llvm-23.1.0-16696adc-prmt.json:69847316bd21a159daed299392114df5d9596faa93d5a2dfee9e12003f7b20cc",
-      "rust-llvm-23.1.0-16696adc-redux-f32.json:54a42cc223a9c0bae1638c89f2327e6cc889b0cd046be8f406fed9a809e37ee7",
-      "rust-llvm-23.1.0-16696adc-register-mma-ampere-float.json:f12382d3cb8ec45337311206324f35d76051dab36f6973643ce56478c2d060e5",
-      "rust-llvm-23.1.0-16696adc-register-mma-b1.json:629f2dd9244ce6a58c09e39ceac4f62c9d3b054a108447eeee0f82a29d7d524f",
-      "rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f16.json:f8f165a733b58fe84dd397a7113ff79cac046c8417ca2702eb4384c8289e9f3c",
-      "rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f32.json:a32d0b8c44f3cc90516dc9ccb838c6fa3dbdb3b41f0906c17cdb994f3dcb0770",
-      "rust-llvm-23.1.0-16696adc-register-mma-fp8.json:6a4e511a694d3e2c77f0404e016471a73311a4df4cf5983f9a23fe413690c2d0",
-      "rust-llvm-23.1.0-16696adc-register-mma-int4-m16.json:b42be6636312468798ef1f7e5aecb2d4243a8556d28ed1c32e4bc0e183f3dc5f",
-      "rust-llvm-23.1.0-16696adc-register-mma-int4-m8.json:ba325db8a87fec566f495d5610c4c74ce9bb41a3d0e3303148907645cdfc9842",
-      "rust-llvm-23.1.0-16696adc-register-mma-int8-m8.json:8cd264dfb5a1e66092f4bc56a1b95feefe50883f482724ae641ff6dcd2c6a3a0",
-      "rust-llvm-23.1.0-16696adc-register-mma-int8.json:bae39a189e006626d12a540b16669cea5ba9c91ff84b5cfabf6d92a22f71476e",
-      "rust-llvm-23.1.0-16696adc-register-mma-mxf8f6f4-f32.json:2afcec08ecec8332a8f78ec9649ab7f7ba1e818fa91cfab4df43824651a67616",
-      "rust-llvm-23.1.0-16696adc-register-mma.json:33d5d636dd9c3453f01674f047e1b562ac347cca25233e133182cdc1fd9c35fb",
-      "rust-llvm-23.1.0-16696adc-scalar-arithmetic.json:6b8c10d5e6809316cdc0b4bd424b30ac8ab48006c60153488d68daf68de50188",
-      "rust-llvm-23.1.0-16696adc-scalar-conversion.json:d1b2271f40a573ac62e474a62c83a20154b0001e3603571bc81246168da13d65",
-      "rust-llvm-23.1.0-16696adc-scalar-math.json:1a9e7994e07b5b53390df11a7e1805aa46176d21c8ed7e988433aee578c4dd47",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-fp8.json:a5e3432308b2e8e937584591653e3f4cb10a3a0f51a50c20a03d2d4492c36e30",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-int8-m16n8k64.json:2c0588c11569a4057647c099cfca9adfc449f3b3aea4e387f440a47e79d9b241",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-int8.json:9023c5c47bf1ad7266a13be8c4024d34e4e7d1efed6582affbbb9cdcdcfbea01",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-ampere-float.json:c4356a2f779613e81041f257a2f976f3edf302be5b0445eab8888841cc89361f",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f16.json:8569d7303b1a5af418214f2548e4b10c825c06c524dbecbc35b8ab0d4e2cf7ba",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f32.json:d6b99f5f4fb88b145f24dba9da662e052137b05a6d2616731d08ab858a4e0541",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k128.json:3a65cb8976f0fb35ca50a7e58943ed092d4e33a8dba84b637402bd3224d7ef21",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k64.json:c670ca442ab1a2373a8029bdc6c567d5df2e1af6df6bd268b1646373f16faf5e",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8-m16n8k64.json:c4bc5eb31670b5be47c4f62f73c202455b5b2fe85544f3195151d2051cf85e09",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8.json:3beb54235f31aec145a1db0b7405fa2a14f99f902c231ff33f45bc276e6794aa",
-      "rust-llvm-23.1.0-16696adc-sparse-mma-standard-int4.json:921cdd8ea3aee8e00811161d5b8aac1adfc8e660a130a49c9dac6d6ee948c56c",
-      "rust-llvm-23.1.0-16696adc-special-registers.json:767656ba6449cd0707f972fd688bb0f170d4d418aacd130209ea734e7eab24a7",
-      "rust-llvm-23.1.0-16696adc-stmatrix.json:ba2f4d6b8b361a6d052395c749295f9ad2174047322b9ac022764c5a288f05e9",
-      "rust-llvm-23.1.0-16696adc-tcgen05-control.json:93eec293a47d692a92fb3ad057e787e31b0e25507cbebdbb4b7dcbed35056c11",
-      "rust-llvm-23.1.0-16696adc-tcgen05-cp.json:2709ecd0983ab4efb36cfe24b1bd277369510ae3f98b337c4b73d715b338e4a9",
-      "rust-llvm-23.1.0-16696adc-tcgen05-ld.json:5dc2b3df8abdfb79ce0a2ae41a286f8d8d6eb99a584353bc884fc1d5b1857e3d",
-      "rust-llvm-23.1.0-16696adc-tcgen05-ldst-offset.json:039b67da5f9edcfc80454b55f3a147880c9d96d56f64de2d82f2070c94a80f33",
-      "rust-llvm-23.1.0-16696adc-tcgen05-mma.json:8bb4d7fe8ecc7815e904eb1f41ce3ccc670d946cdcda567b188e64e7a4d7e7ba",
-      "rust-llvm-23.1.0-16696adc-tcgen05-st.json:3ac8a0870333ea8a4b93a1f3ea95b27fb76477b3c8e2597b3df549125a7de010",
-      "rust-llvm-23.1.0-16696adc-tcgen05.json:9a7fda406f9f8204ffcb99f5d055550b64fa34e939b64c2ae65a9249f7f3773b",
-      "rust-llvm-23.1.0-16696adc-threadfence.json:35e804e8bb9c2bd2358046df50c4ccee779fbe3c0290742af0c213fbb16c2aea",
-      "rust-llvm-23.1.0-16696adc-tma-descriptor-control.json:6b9255dd80bb30b3b25fea0bf00a9abdfdfa8d18926a51dee38e5133e6ee3a3f",
-      "rust-llvm-23.1.0-16696adc-tma-reduce.json:e58a0d70e1678bb1a95b0b455b7fc424c8a17e6ec7f2bc2c4c8f1c51c034cb16",
-      "rust-llvm-23.1.0-16696adc-tma.json:2a28f49b6965d82324b82f510e2b9a39df775aa1bb05a0c7d8dd6ba25d4719da",
-      "rust-llvm-23.1.0-16696adc-wgmma-control.json:2993e8ebfbbc966a41142675ce97dfbf421642f082c8a9457217f2fdd5690020",
-      "rust-llvm-23.1.0-16696adc.json:20731ccd24a22ba6aa8d2d4aa81e6325bf97c8916423c895c1b369b470090a7e"
+      "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix.json:d97aa88300a7b6cf355711ce82d1a719ba27c1f8dce4d94a746ee664973dd309",
+      "rust-llvm-23.1.3-aab11ee0-clc.json:5eb99c01287207b78db2a7cf3b2b60d932a64377120e65b39121f4bb65028061",
+      "rust-llvm-23.1.3-aab11ee0-cluster-barrier.json:2a32bd2e2e58ac8f48fc9eb590c44985ac9ac5ffefb62374f84138d267dfb824",
+      "rust-llvm-23.1.3-aab11ee0-cluster-memory.json:bf24c8a01ce42d94ae48fe814cff5d6e2c6967f8036a1c22dadfac82699cf336",
+      "rust-llvm-23.1.3-aab11ee0-cluster-sreg.json:f69a0fc34bd655fb322d7e131aeb148ea0a232eac6b7510ef06ac079acf1a5ca",
+      "rust-llvm-23.1.3-aab11ee0-cp-async-mbarrier.json:262e59cf2d752943aa1ffc99b2f8ccda965162a6f8726ed226873d8b174d0659",
+      "rust-llvm-23.1.3-aab11ee0-cp-async.json:6979a2d0eeec714e0d739ba713ea5d55f3532aaf5d05b3a2c866e6b40defdbea",
+      "rust-llvm-23.1.3-aab11ee0-debug-control.json:81e4d6aeca03af681ab23918368fd7aa848e2637c459b48cfb5421a631d824d6",
+      "rust-llvm-23.1.3-aab11ee0-elect.json:daf7b6f024a8e64220adaca12427dd35b3b2c4a0db87205c089794b27a067726",
+      "rust-llvm-23.1.3-aab11ee0-execution-control.json:bbcb134baeae611ee0c8c86ee2b67a3d2b91645c32c9f6cc9283174dc7a72136",
+      "rust-llvm-23.1.3-aab11ee0-extended-minmax.json:913d001130f4b04651eff4b0db27a50628f3834a0bff5d823f018217bf22f2da",
+      "rust-llvm-23.1.3-aab11ee0-f32x2.json:fa65485356ead2102359249b9c41b3dacd75fdb113b9bb4659a70e3b212e1433",
+      "rust-llvm-23.1.3-aab11ee0-integer-minmax.json:94bfd0d84bf57737acda19278a10cb566c25677c5688e6a918c83da822100480",
+      "rust-llvm-23.1.3-aab11ee0-mbarrier-basic.json:193966219733100689a43dfde7e9f8aba1f3e3ba2cf80b14bd4893504f900489",
+      "rust-llvm-23.1.3-aab11ee0-mbarrier-extended.json:c92addb73ebd8ef4f3bb2d5902a9e7f280ba898d10b54682d44f26cd2968fcd8",
+      "rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8-f16x2.json:20124138104fe9f71a583a73cb8656c7d91aa10366e9e041499e024578c1a2e3",
+      "rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8.json:42561b5ee707545ccabe2b3b78de12a31614893611e0deccbb498a3ba2967fd1",
+      "rust-llvm-23.1.3-aab11ee0-prmt.json:c77aaaa7fed2d55c2319e286f092bc440e34ef6e6aaf08abe712a3179745d990",
+      "rust-llvm-23.1.3-aab11ee0-redux-f32.json:f02c6ee175ddd347d33217fb2f72e0823c202e2cf1992646918ef61fbabfac0d",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-ampere-float.json:2f31c8baffe053dc086b15cf304a3b1f50196fc349d642525529d4f7442d53a9",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-b1.json:5411cbdc4b090f00b71bdeda70742f4f32fd965057241750ed3ffd2a80301b04",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f16.json:9ffae2d4518f432526fe4f5435bce3d165ee636930443f311a6401499f67a334",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f32.json:3428432434f81ab26a325d5d62a041dc8397631335c8a39efcecc926d61fc4df",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-fp8.json:7e43e0eeb70e982dc13c743a3049b9261644d7570810c2261a74f7a65463973e",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-int4-m16.json:922f998fb2e65cf15193eb20ea1d2fef2b24390eea68ce273813d657faa62f62",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-int4-m8.json:79cc39c9713df2dda432ce7c3008ad671f220184125dca6b3d169884029b1490",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-int8-m8.json:cad96d998bb1d6c4bfd3b99e9b89f09dbcf2c736db0e0e72a75e92c48887431c",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-int8.json:7b30ec02a709517c4e2c8158044e26710a9a0362b91e2ef09856292f9c18541c",
+      "rust-llvm-23.1.3-aab11ee0-register-mma-mxf8f6f4-f32.json:74930317085a35821ff129a56351901c784e07d28fdc8b89cae608dc32e95802",
+      "rust-llvm-23.1.3-aab11ee0-register-mma.json:85e30e8e7c3db6ab6e14cf364e16ca761bcca123e8fcf378080b82cd8c025f75",
+      "rust-llvm-23.1.3-aab11ee0-scalar-arithmetic.json:2b75b6fabd4baec416c2836d54d0ec9f2d2f0381f6a22e0b6b923703c729419f",
+      "rust-llvm-23.1.3-aab11ee0-scalar-conversion.json:fc454571fa8ee04649455963022b7a3ab4eaecbdffa561522b00f495c8153d78",
+      "rust-llvm-23.1.3-aab11ee0-scalar-math.json:f1d1d54c5cb53cbef1fc0cbd447c18b9a9c9909405174d8f23576e0e4db1bdbf",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-fp8.json:f21f3182c4d55d51ca4e775ecf7b68abb577d31592ebaa53c11de6afdf8e173e",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-int8-m16n8k64.json:dc80ab1a5f91024c3926ee08a18292b468a5134166bef96d65084e50e98c023a",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-int8.json:261b4e8a235c5b1d684e03cde8fc60b7976ce2e0b480fc9eb326ae1f1529afa9",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-ampere-float.json:6c80f4eef1739f4d90a63ed29a1b1c977dda7f1eeb4876a2d2119585f8f6c871",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f16.json:2e5d345e65c47dad346ccd24e935b2ce5368c8e8b7da8703be9f7f98b3a7838a",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f32.json:50729befbd2a7423dd6c9a4ba169167671fe7925330ca7c3d2c971c527dd2ced",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k128.json:cecc9286beefd58d0fe6e2e1ec9a236c7442ed4df10573f4db6499514830234c",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k64.json:0aeaa46491ba599550c1af9fcba5170bdd1945ebcc8a1bbe219211d37380ded0",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8-m16n8k64.json:c81dc82bba56022ebb7f72ef227ff53f62160551f124ecdf4cb26504947da846",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8.json:d65a25ea4a17a217f93a487799f15e555a066d171621f9e5072206b36033418b",
+      "rust-llvm-23.1.3-aab11ee0-sparse-mma-standard-int4.json:cecf5bca2adb10b65a5d01c29dd0ef0482437c20bce183bf970d90a95445de82",
+      "rust-llvm-23.1.3-aab11ee0-special-registers.json:cacdf70d6e739898f5bda6d0d3f65d1a9112740898dd4ccd31a252bbd6e60df9",
+      "rust-llvm-23.1.3-aab11ee0-stmatrix.json:dc7b371a047fd47aba9bdc153a7021cbbc153312d87c0ef0eb3c9fd25c857b42",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05-control.json:6f8b15891a5265a3e248645c76d06a67fcd72358a682a4ce6a563c2c3808feae",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05-cp.json:8c64f026a112c49654dbc76b8157b3553621dc97ad9780157b4f6bbbce559260",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05-ld.json:a834e8c9f1c2c9e81ea3b5e8954998b54682da729f55d1c7101445e33fcd1ec9",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05-ldst-offset.json:e4c5993a5b6768e626ca719eb58de3b7c4f9b09844691f732737197647643c04",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05-mma.json:19f80f2aac86124655819c94f3242fac17a5045e2bd7d532dd0a383cd0e06081",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05-st.json:e559aa775e3e964f9889483e9d9d8c45b46628931ff78050c46f33e448f69e86",
+      "rust-llvm-23.1.3-aab11ee0-tcgen05.json:b6cf11d81c0880877ac5ad8958231222c728d5ab0243ddc84f1db901fbc8608b",
+      "rust-llvm-23.1.3-aab11ee0-threadfence.json:087343e0247e7f9571c74fe763c0d48fbd26e246a21a9f10e5466a9a4457f83e",
+      "rust-llvm-23.1.3-aab11ee0-tma-descriptor-control.json:7b6deaedfb8f8ac4b6551c47701cdefd04867431ac4dc5c3fbaf8f8bc604c9e9",
+      "rust-llvm-23.1.3-aab11ee0-tma-reduce.json:e472b96476cad1d406fdac6dd8209645357f068cbd545669e2f09a8c6a232b79",
+      "rust-llvm-23.1.3-aab11ee0-tma.json:97fb372aa7c6966bd01667da32d9d0857e385776131cb548f7d70f73b52718bc",
+      "rust-llvm-23.1.3-aab11ee0-wgmma-control.json:23a906f42faddd256ed0b36ff24c6dd8071a819776f67bb64a0939bae59aed00",
+      "rust-llvm-23.1.3-aab11ee0.json:0ddaa5355b5289f2ca822e2f06dc1755108cf6804180cae25a596d36f99f186b"
     ]
   },
   "intrinsics": [
@@ -223,9 +223,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -235,7 +235,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -248,8 +248,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -259,11 +259,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX abs.bf16x2 with =r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact abs.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fc97be133b128891fbdca9f6a4b9769cfd2712fa35ee81884e52b9b436f6ef32.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -274,15 +274,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbafd197892b5b499ef62752012ccb5e86318641dd9784942872a4ca7d685893.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -452,9 +452,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -464,7 +464,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -477,8 +477,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -488,11 +488,11 @@
                 "sm_70",
                 "ptx65"
               ],
-              "representation": "pure inline PTX abs.f16x2 with =r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact abs.f16x2 inline PTX at PTX 6.5/sm_70. An independent typed call to the resolved llvm.nvvm.fabs.v2f16 symbol emitted the same instruction at PTX 6.5/sm_70 and at its native sm_53 floor; sm_52 failed exact intrinsic selection.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7282ee206f4fdb3a54d1fe47412d9ab41a11954358a8e0ee203f434988aa5191.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -503,15 +503,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX 6.5 emitted by the sm_70 abs.f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 737-byte PTX module at sm_75 and produced a 3,112-byte cubin (SHA-256 24150b058e7b7ad99ffeacd77588a2f1febdf5479f93f966ac0192c9834babf7). SASS realizes packed absolute value as HADD2 with an absolute source. The PTX 6.4 negative module was rejected because abs.f16/abs.f16x2 requires PTX 6.5 or later.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93adc93cea87ea659d8a0b4542f44e1d897cb4c1cd53a06dc08becdac7de7ea6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -675,9 +675,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-activemask"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -687,7 +687,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.2",
             "hardware": {
@@ -700,8 +700,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -711,11 +711,11 @@
                 "sm_30",
                 "ptx62"
               ],
-              "representation": "typed llvm.nvvm.activemask() declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, side-effect, and inaccessible-memory contract.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a17f68d96e936c632dbeb06ea74eae28b193c05cdfdc08373f0715f348eacf1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -726,11 +726,11 @@
                 "sm_30",
                 "ptx62"
               ],
-              "representation": "typed zero-operand activemask call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted activemask.b32 at the imported instruction floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 246910573ff59e4feb1a3b8880eb69d36c7a14f49debfe41f1207db7b4ea51d4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -873,9 +873,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -885,7 +885,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -898,8 +898,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -909,11 +909,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "pure inline PTX add.rn.bf16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact add.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2c103ab626814fb233a6e15e90a48c2a6a5b78f7f13e38c4c8c6cebca28ef0d0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -924,15 +924,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "PTX emitted by the combined three-form sm_90 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af1befea9f33a173ed1a34fa549dfd6e8cf85f450042ec4948cdcd9eef20051f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -1080,9 +1080,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -1092,7 +1092,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -1105,8 +1105,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -1116,11 +1116,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX add.rn.f16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b7f49dda4585f12ddfc9ecaf71517120de1aefd78a9f843fcb511f81bbabaf4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -1131,15 +1131,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b33cc169935b856e4ff58245496b889d56c48c772653634f963facb395c24326. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -1287,9 +1287,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -1299,7 +1299,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -1312,8 +1312,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -1323,11 +1323,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -1338,15 +1338,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -1494,9 +1494,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -1506,7 +1506,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -1519,8 +1519,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -1530,11 +1530,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -1545,15 +1545,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -1728,9 +1728,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -1740,7 +1740,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -1753,8 +1753,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -1764,11 +1764,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -1779,11 +1779,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -1794,15 +1794,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -1993,9 +1993,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -2005,7 +2005,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -2018,8 +2018,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -2029,11 +2029,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2044,11 +2044,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2059,15 +2059,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -2258,9 +2258,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -2270,7 +2270,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -2283,8 +2283,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -2294,11 +2294,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2309,11 +2309,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2324,15 +2324,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -2524,9 +2524,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -2536,7 +2536,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -2549,8 +2549,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -2560,11 +2560,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2575,11 +2575,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2590,15 +2590,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -2791,9 +2791,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -2803,7 +2803,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -2816,8 +2816,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -2827,11 +2827,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2842,11 +2842,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -2857,15 +2857,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -3057,9 +3057,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -3069,7 +3069,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -3082,8 +3082,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -3093,11 +3093,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3108,11 +3108,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3123,15 +3123,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -3322,9 +3322,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -3334,7 +3334,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -3347,8 +3347,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -3358,11 +3358,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3373,11 +3373,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3388,15 +3388,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -3587,9 +3587,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -3599,7 +3599,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -3612,8 +3612,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -3623,11 +3623,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3638,11 +3638,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3653,15 +3653,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -3853,9 +3853,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -3865,7 +3865,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -3878,8 +3878,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -3889,11 +3889,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3904,11 +3904,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -3919,15 +3919,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -4120,9 +4120,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -4132,7 +4132,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -4145,8 +4145,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -4156,11 +4156,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4171,11 +4171,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4186,15 +4186,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -4386,9 +4386,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -4398,7 +4398,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -4411,8 +4411,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -4422,11 +4422,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4437,11 +4437,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4452,15 +4452,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -4651,9 +4651,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -4663,7 +4663,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -4676,8 +4676,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -4687,11 +4687,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4702,11 +4702,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4717,15 +4717,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -4916,9 +4916,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -4928,7 +4928,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -4941,8 +4941,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -4952,11 +4952,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4967,11 +4967,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -4982,15 +4982,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -5182,9 +5182,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -5194,7 +5194,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -5207,8 +5207,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -5218,11 +5218,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -5233,11 +5233,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -5248,15 +5248,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -5449,9 +5449,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -5461,7 +5461,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -5474,8 +5474,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -5485,11 +5485,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -5500,11 +5500,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -5515,15 +5515,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -5715,9 +5715,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -5727,7 +5727,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -5740,8 +5740,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -5751,11 +5751,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -5766,11 +5766,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -5781,15 +5781,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -5980,9 +5980,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -5992,7 +5992,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -6005,8 +6005,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -6016,11 +6016,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6031,11 +6031,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6046,15 +6046,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -6245,9 +6245,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -6257,7 +6257,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -6270,8 +6270,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -6281,11 +6281,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6296,11 +6296,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed llvm.nvvm.add calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6311,15 +6311,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -6511,9 +6511,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -6523,7 +6523,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -6536,8 +6536,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -6547,11 +6547,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6562,11 +6562,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6577,15 +6577,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -6778,9 +6778,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -6790,7 +6790,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -6803,8 +6803,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -6814,11 +6814,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6829,11 +6829,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -6844,15 +6844,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -7054,9 +7054,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -7189,9 +7189,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -7324,9 +7324,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -7437,9 +7437,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -7449,7 +7449,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -7462,8 +7462,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -7472,11 +7472,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed declarations for the four PTX 7.8 cluster-barrier records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -7487,11 +7487,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "one typed probe retaining the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -7502,15 +7502,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX containing the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -7691,9 +7691,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -7703,7 +7703,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -7716,8 +7716,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -7726,11 +7726,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed declarations for the four PTX 7.8 cluster-barrier records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -7741,11 +7741,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "one typed probe retaining the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -7756,15 +7756,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX containing the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -7946,9 +7946,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -7958,7 +7958,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -7971,8 +7971,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -7981,11 +7981,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed declarations for the two PTX 8.0 relaxed cluster-barrier records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 2e2c065b1abe60a62a61678dca92352084f523640ebf0feb0dd47e0f8417b18f and bitcode SHA-256 7522f3d31dd29407d8c5d8b4c35ad5f38f663067ad3c415effb5d096b4d93463.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c5b7ef9743a0c36a3ad27cf3093c5d3160111a0acee880922a93dab4ce330a7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -7996,11 +7996,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "one typed probe retaining the two PTX 8.0 relaxed cluster-barrier instructions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted each exact instruction once; PTX SHA-256 4e8650939b44a80b88fe91ede1a9568ac5526889254c9d6954d52176c10a9714.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 462bdc8a89623f986fa9c307a069d444805dcb8644ef7f9793f45a48a5cc94c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8011,15 +8011,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX containing the two PTX 8.0 relaxed cluster-barrier instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 698cf65618fd422030be4645284d6ab56b24f75d0e955a49380ef78bb8e044a8.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5501565e9a498f7f5dba887004bc2cf506c57249b8a4a2e8fdebcefc8007b146.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -8201,9 +8201,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -8213,7 +8213,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -8226,8 +8226,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -8236,11 +8236,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed declarations for the two PTX 8.0 relaxed cluster-barrier records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 2e2c065b1abe60a62a61678dca92352084f523640ebf0feb0dd47e0f8417b18f and bitcode SHA-256 7522f3d31dd29407d8c5d8b4c35ad5f38f663067ad3c415effb5d096b4d93463.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c5b7ef9743a0c36a3ad27cf3093c5d3160111a0acee880922a93dab4ce330a7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8251,11 +8251,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "one typed probe retaining the two PTX 8.0 relaxed cluster-barrier instructions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted each exact instruction once; PTX SHA-256 4e8650939b44a80b88fe91ede1a9568ac5526889254c9d6954d52176c10a9714.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 462bdc8a89623f986fa9c307a069d444805dcb8644ef7f9793f45a48a5cc94c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8266,15 +8266,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX containing the two PTX 8.0 relaxed cluster-barrier instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 698cf65618fd422030be4645284d6ab56b24f75d0e955a49380ef78bb8e044a8.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5501565e9a498f7f5dba887004bc2cf506c57249b8a4a2e8fdebcefc8007b146.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -8457,9 +8457,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -8469,7 +8469,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -8482,8 +8482,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -8492,11 +8492,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed declarations for the four PTX 7.8 cluster-barrier records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8507,11 +8507,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "one typed probe retaining the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8522,15 +8522,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX containing the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -8711,9 +8711,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -8723,7 +8723,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -8736,8 +8736,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -8746,11 +8746,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed declarations for the four PTX 7.8 cluster-barrier records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8761,11 +8761,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "one typed probe retaining the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -8776,15 +8776,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX containing the four PTX 7.8 cluster-barrier instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -8995,9 +8995,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -9007,7 +9007,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -9020,8 +9020,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -9030,11 +9030,11 @@
               "targets": [
                 "sm_30"
               ],
-              "representation": "typed counted-barrier arrival declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 6844935a6b2fc799ecf3245780589f5de02583685caae0267be1f37668535a2b, bitcode SHA-256 d014888592c0d71d5c569711143349e9daaea3079578144385a41fa055e87992, and canonical IR SHA-256 e2a2a5029602f80fb55f6f90e96083d51ce76da3c019b666aad025d0e64c0987.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 70c603a8e3092012e74c6e6f5a2330489176c4af38499c379753f095ae899a2a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9045,11 +9045,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "four register/immediate forms of barrier.arrive",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 94113e81da1085d1688c37f811dc497e0963be3af0f703f03862268e4a14678f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6a153e4c537c56cc43402bcf7786c6540c696ff807919d11973d650c7bc9a78c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9060,15 +9060,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "LLVM-produced PTX containing the four barrier.arrive forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9499277c4334862a892ac0c79eb49c79fc5c0dc889639a4b22dd5d0ea2134d0c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -9241,9 +9241,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -9253,7 +9253,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -9266,8 +9266,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -9276,11 +9276,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed aligned counted-barrier arrival declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 042ab4e18a0df19c17f139123fd21759a9b179221f8d303f9d13c8079484f5cf, bitcode SHA-256 8c4a5ec77772763a7deb91899e5ae3a1fd10df067d619b985b4e6e971e8b2319, and canonical IR SHA-256 6ce3356d92a9740dfeb0f53cf4aa551586a1ee201a2210dadf631131e26ce047.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 171894b0dd863526a8ff4cf86ecc05f04a47032b49254ae9d5a03f58bcc5eaf4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9291,11 +9291,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "four register/immediate forms of bar.arrive",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 6b4d17807a12f8298fca5f4957b8e3da5ed27fcbd47111f8a173bbe61d5c8aad.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cb96f04a02a2ae0cde2c1d3691ca3d2752577a6b32337b39ab1afecaa7f98249.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9306,15 +9306,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX containing the four bar.arrive forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11118e20cc0d2ec90152a19b6e75a9dbcc4f09a1d5698c030ebf890c88095fa1. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -9495,9 +9495,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -9507,7 +9507,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -9520,8 +9520,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -9530,11 +9530,11 @@
               "targets": [
                 "sm_30"
               ],
-              "representation": "typed counted-barrier declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 73bcf70a8647d141dea6d68aba4ab626818aef73c25eaf85b9c415814dcf0e71, bitcode SHA-256 50b842e412973300a1b0c67190c173f7f5f744db15efc9d639daa3b5c7ec33c6, and canonical IR SHA-256 d068e7a4ff32510dd439f5e92e97374a319e4ba6c1a6380cff5c855a0cde6a3f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1ec32a6b202b53ac34f270d293fd60d90d5d6de4dacef2b7d7912ab16698c69f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9545,11 +9545,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "four register/immediate forms of barrier.sync",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 b220ca5fdae82f6048cc1e930fa07a9f6e8cf23a610476f0cf469a01e985a85d.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa2ed67b2503e249a9ae63ef2a7436c41d098c27956fe0faf441c0e961db4151.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9560,15 +9560,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "LLVM-produced PTX containing the four barrier.sync forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a9e71fa97f369891143a3c32fa8c80f1c28be09176445e1edfff81c54615ffd. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -9741,9 +9741,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -9753,7 +9753,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -9766,8 +9766,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -9776,11 +9776,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed aligned counted-barrier declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 f5faf9f6ac1de5de821476c35d6cfed39ffb9abc1de2e7ac9038c6f51c7bda6c, bitcode SHA-256 3c90f0f9686014d1074c753f644ca9dac74397f465cb78f23e726820af47345d, and canonical IR SHA-256 70ee92254f4aa29db9731a3beddef77865f6f5bff7a81a7c66aa076726076ea4.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c313db28069767d98c8d477dd5512e1d4dae29dc52ea0d691fe03edb03c5e50.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9791,11 +9791,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "four register/immediate forms of bar.sync",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 d3fbf9d7aa025ee2f4836094b8652fb29a5557e94bbc73689c5b18dd04edc47a.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0615f66795bc5a379d6cf2426d83a39d4baae46194d61b6cde902e9b8c8ab367.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -9806,15 +9806,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX containing the four bar.sync forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 999e27bbd3d4ccd52c8f1366731b70ab35d9d67ed19125ffb66ac7bf160b4bbc. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -9967,9 +9967,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -10070,9 +10070,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -10173,9 +10173,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -10276,9 +10276,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -10379,9 +10379,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -10482,9 +10482,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -10562,9 +10562,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-brkpt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -10574,7 +10574,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-debug-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-debug-control",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -10587,8 +10587,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -10598,11 +10598,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one LLVM IR probe with side-effect inline PTX for trap, brkpt, and pmevent 15",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -10613,15 +10613,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "1,011-byte LLVM-produced PTX module containing trap, brkpt, and pmevent 15",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -10772,9 +10772,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -10784,7 +10784,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-clc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-clc",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -10797,8 +10797,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -10807,11 +10807,11 @@
               "targets": [
                 "sm_100"
               ],
-              "representation": "typed declarations for unicast try-cancel and all four query operations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -10822,11 +10822,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "one typed LLVM kernel retaining unicast try-cancel and all four CLC queries",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -10837,15 +10837,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM PTX retaining the unicast and query kernel",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -11023,9 +11023,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -11035,7 +11035,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-clc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-clc",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -11048,8 +11048,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -11058,11 +11058,11 @@
               "targets": [
                 "sm_100"
               ],
-              "representation": "typed declarations for unicast try-cancel and all four query operations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11073,11 +11073,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "one typed LLVM kernel retaining unicast try-cancel and all four CLC queries",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11088,15 +11088,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM PTX retaining the unicast and query kernel",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -11274,9 +11274,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -11286,7 +11286,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-clc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-clc",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -11299,8 +11299,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -11309,11 +11309,11 @@
               "targets": [
                 "sm_100"
               ],
-              "representation": "typed declarations for unicast try-cancel and all four query operations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11324,11 +11324,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "one typed LLVM kernel retaining unicast try-cancel and all four CLC queries",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11339,15 +11339,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM PTX retaining the unicast and query kernel",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -11525,9 +11525,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -11537,7 +11537,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-clc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-clc",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -11550,8 +11550,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -11560,11 +11560,11 @@
               "targets": [
                 "sm_100"
               ],
-              "representation": "typed declarations for unicast try-cancel and all four query operations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11575,11 +11575,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "one typed LLVM kernel retaining unicast try-cancel and all four CLC queries",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11590,15 +11590,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM PTX retaining the unicast and query kernel",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -11782,9 +11782,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -11794,7 +11794,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-clc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-clc",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -11807,8 +11807,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -11817,11 +11817,11 @@
               "targets": [
                 "sm_100"
               ],
-              "representation": "typed declarations for unicast try-cancel and all four query operations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11832,11 +11832,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "one typed LLVM kernel retaining unicast try-cancel and all four CLC queries",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -11847,15 +11847,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM PTX retaining the unicast and query kernel",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -12058,9 +12058,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -12070,7 +12070,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-clc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-clc",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -12103,8 +12103,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -12113,11 +12113,11 @@
               "targets": [
                 "sm_100a"
               ],
-              "representation": "typed declaration for multicast try-cancel",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 390-byte source with SHA-256 e4ed73cd8c39af86e8723ca03126c44f31cc9511a73b6ece710b7875bc128b14 and produced 1,912-byte bitcode with SHA-256 2d37df5745e6221f4b75791540de98edab835e3370f36043bcf6aa7af4a3c22d.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd29e45eadad739b58e43893296b4b8ac5e05d43c3c3298836fb126503d61e83.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -12128,11 +12128,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one typed LLVM kernel retaining multicast try-cancel",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM 22 emitted one exact multicast instruction with no extern call for sm_100a/PTX 8.6, sm_101a/PTX 8.6, sm_103a/PTX 8.8, sm_110a/PTX 9.0, sm_120a/PTX 8.7, and sm_121a/PTX 9.0. Their PTX SHA-256 values were 366b5155464356e2e01d29fb3ad209f365ba68fc2e540489ce7e2f5830658cde, bf925481f00d8023b5493cf2daee655ba90e32486d16b140c013df87bc1b5625, 6310890cf30d93ae0d02d63362350bfd13428ec4522358d80b459a3b1e86435b, 7ae967df1894e5a0158eabed03a607eeb971d1d2f7e5ad2892ca7f632d95ccd7, a4d5dd50691a5e0d3a32d4c77bf87258009ecebf2d66d00bb73292d0cbb4fa4e, and c52b501ba903c302697c994c9ea3abd4506b8a3aee1fd6ca320cad533140c894. Plain sm_100 failed instruction selection, as required.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 423445d4b9994ee20be92cbabc0ad7c8ea667c6360848df3b2a5be7020df4051.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -12143,15 +12143,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "typed LLVM PTX retaining the multicast kernel",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas assembled the exact probes for sm_100a, sm_103a, sm_110a, sm_120a, and sm_121a. Their cubin SHA-256 values were 30160dcbe4de4c64e9b78f907e2663cdfb244fbe1a5dfbba66a6b8a1c52d5d76, 357d54085a3c661f8b0841d933d2d1600a688b54a5004ac0610469c35d016c57, 6d3afe7f862c5f2b5931972aa5c5dac9e2649f14407dfe67ade547b72af951db, e481feac9e4382025b1204c38614a1f5bdf141f9ce226da45d4ff835d58fdaba, and dcb515f3f0bb87fb6ffb2e74b78542979a0264f56b6df43a3913786e201c44f9. CUDA 13.3 does not expose the LLVM-only sm_101a name.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f8543fdf61a5a323f880af2fdfe1b86d2ee1311a024f5a9e88d6c40a23845925.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -12345,9 +12345,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clock-clock-hi"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -12357,7 +12357,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -12370,8 +12370,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -12380,11 +12380,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -12395,11 +12395,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed calls lowering to six exact special-register reads",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -12410,15 +12410,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -12596,9 +12596,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clock64"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -12608,7 +12608,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -12621,8 +12621,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -12631,11 +12631,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -12646,11 +12646,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed calls lowering to six exact special-register reads",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -12661,15 +12661,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -12846,9 +12846,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctarank"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -12957,9 +12957,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13068,9 +13068,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13179,9 +13179,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13284,9 +13284,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctarank"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13395,9 +13395,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13506,9 +13506,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13617,9 +13617,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13728,9 +13728,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13839,9 +13839,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -13950,9 +13950,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -14061,9 +14061,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -14172,9 +14172,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -14283,9 +14283,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -14386,9 +14386,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-cos"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -14398,7 +14398,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -14411,8 +14411,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -14422,11 +14422,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -14437,11 +14437,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -14452,15 +14452,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -14636,9 +14636,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-cos"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -14648,7 +14648,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -14661,8 +14661,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -14672,11 +14672,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -14687,11 +14687,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -14702,15 +14702,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -14883,9 +14883,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -14895,7 +14895,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -14908,8 +14908,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -14918,11 +14918,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -14933,11 +14933,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -14948,15 +14948,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -15128,9 +15128,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -15140,7 +15140,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -15153,8 +15153,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -15163,11 +15163,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15178,11 +15178,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15193,15 +15193,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -15384,9 +15384,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -15396,7 +15396,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -15409,8 +15409,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -15419,11 +15419,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15434,11 +15434,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15449,15 +15449,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -15645,9 +15645,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -15657,7 +15657,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -15670,8 +15670,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -15680,11 +15680,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15695,11 +15695,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15710,15 +15710,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -15904,9 +15904,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -15916,7 +15916,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -15929,8 +15929,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -15939,11 +15939,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15954,11 +15954,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -15969,15 +15969,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -16168,9 +16168,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -16180,7 +16180,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -16193,8 +16193,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -16203,11 +16203,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -16218,11 +16218,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -16233,15 +16233,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -16430,9 +16430,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -16442,7 +16442,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -16455,8 +16455,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -16465,11 +16465,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -16480,11 +16480,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -16495,15 +16495,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -16697,9 +16697,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -16709,7 +16709,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -16722,8 +16722,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -16732,11 +16732,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -16747,11 +16747,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -16762,15 +16762,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -16962,9 +16962,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -16974,7 +16974,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -16987,8 +16987,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -16997,11 +16997,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17012,11 +17012,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17027,15 +17027,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -17232,9 +17232,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -17244,7 +17244,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -17257,8 +17257,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -17267,11 +17267,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17282,11 +17282,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17297,15 +17297,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -17500,9 +17500,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -17512,7 +17512,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -17525,8 +17525,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -17535,11 +17535,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17550,11 +17550,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17565,15 +17565,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -17781,9 +17781,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -17793,7 +17793,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -17818,8 +17818,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -17828,11 +17828,11 @@
               "targets": [
                 "sm_100a"
               ],
-              "representation": "two individually generated typed gather4 TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized both declarations. SHA-256 over sorted per-operation ID/hash lists is 4c60ef82171ca5e9bdd92a90683367bbb73023593d27e1af329fa9af9cbce511 for source IR, d186a1e64a5832f4bbe70323a18796cfaef6ab5d4b685b32709085ff1c24e960 for bitcode, and 609578ee9ee7d3bc1bb280cdf57aaa6b7bb1ea1a341a159ff3dec47c0d235ec0 for canonical IR.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcf9f4e3b994a7ccfade3699fa2c7393855b07aee83a70d3c449d4f4260098c4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17843,11 +17843,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two typed gather4 TMA prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted both exact instructions at their catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 d5149bf319d478db6e1e40978c5acb62203fbb92fa5d767f0a4ddeaa3eb8fb8e.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 701c7338ba48b42868907f3900d1074b019fab8e257f5676f9a92e368c4718f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -17858,15 +17858,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two LLVM-produced gather4 TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both modules; the sorted per-operation ID/cubin-hash list has SHA-256 3a914c3e352965013e8a3b1a937f4faa4b4966830f04ca63a3c4d982427cf3d5.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0635d30976c9afa0d905c53f55599ce0fa36a009828f00a5ae0f5283857a21de.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -18084,9 +18084,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -18096,7 +18096,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -18121,8 +18121,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -18131,11 +18131,11 @@
               "targets": [
                 "sm_100a"
               ],
-              "representation": "two individually generated typed gather4 TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized both declarations. SHA-256 over sorted per-operation ID/hash lists is 4c60ef82171ca5e9bdd92a90683367bbb73023593d27e1af329fa9af9cbce511 for source IR, d186a1e64a5832f4bbe70323a18796cfaef6ab5d4b685b32709085ff1c24e960 for bitcode, and 609578ee9ee7d3bc1bb280cdf57aaa6b7bb1ea1a341a159ff3dec47c0d235ec0 for canonical IR.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcf9f4e3b994a7ccfade3699fa2c7393855b07aee83a70d3c449d4f4260098c4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18146,11 +18146,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two typed gather4 TMA prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted both exact instructions at their catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 d5149bf319d478db6e1e40978c5acb62203fbb92fa5d767f0a4ddeaa3eb8fb8e.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 701c7338ba48b42868907f3900d1074b019fab8e257f5676f9a92e368c4718f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18161,15 +18161,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two LLVM-produced gather4 TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both modules; the sorted per-operation ID/cubin-hash list has SHA-256 3a914c3e352965013e8a3b1a937f4faa4b4966830f04ca63a3c4d982427cf3d5.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0635d30976c9afa0d905c53f55599ce0fa36a009828f00a5ae0f5283857a21de.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -18385,9 +18385,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -18397,7 +18397,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -18410,8 +18410,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -18420,11 +18420,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18435,11 +18435,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18450,15 +18450,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -18649,9 +18649,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -18661,7 +18661,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -18674,8 +18674,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -18684,11 +18684,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18699,11 +18699,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18714,15 +18714,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -18928,9 +18928,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -18940,7 +18940,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -18953,8 +18953,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -18963,11 +18963,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18978,11 +18978,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -18993,15 +18993,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -19211,9 +19211,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -19223,7 +19223,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -19236,8 +19236,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -19246,11 +19246,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -19261,11 +19261,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -19276,15 +19276,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -19509,9 +19509,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -19521,7 +19521,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -19546,8 +19546,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -19556,11 +19556,11 @@
               "targets": [
                 "sm_100a"
               ],
-              "representation": "one LLVM module containing the CTA-pair TMA declaration and call",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 521-byte source had SHA-256 c8f988c762ae488faa6a7a31f1391a74b33918a088449f4ec2bfbdb1ca33a201; the 1,964-byte bitcode had SHA-256 dd23f8a1a678731e95b7e2283da19530bce758b7b55ded0c98d01e99c971b978.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 065ffbad7cc3f4a868f3de9e7f86d3a4fb1b0ea7047e1ddcdd578192e1fe101c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -19571,11 +19571,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "typed LLVM CTA-pair TMA call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted the exact cta_group::2 instruction with no extern call. The 1,098-byte PTX hashes were 88d30f75f80d7603ec5852572942e3406db97d10a49bdb7b65c64fa1bcea2257 for sm_100a, ed24ee1e914179959d69b7b153d5d65bbead0c864bfeb624871f9c19f5cc17f3 for sm_101a, 451331c3b63021119d0e9ccdc7617e70b505943c69e1066d193ef1b74b54523e for sm_103a, and 13c7be98f933799fafa4055423395bc1d5d4c2f4303e70f27c38ded9a0e1ace8 for sm_110a.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6f52db1ccb491246b76348fb43a73fc961823d087d80432a2cec08dc36439f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -19586,15 +19586,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining the CTA-pair TMA instruction",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubins for sm_100a, sm_103a, and sm_110a with SHA-256 7c522a3420f9856b526e23c6e3dc4067d244b27095057b91ed1b4af27a8abbcf, b3e3934adc407c847b428d93936f7d8e170059f4004218774c8755f71d599419, and ef654bdc632b1887f8622e765f8f5a2d0571b97c28eb39f17fd355483d8b9d31. Separate probes rejected sm_120a and sm_121a because cta_group::2 is unsupported.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eea1fea3e3abbce7a6d377fafd8ad8e2c8fec998a990afbd664b7ce4f3856dff.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -19805,9 +19805,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -19817,7 +19817,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -19830,8 +19830,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -19840,11 +19840,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -19855,11 +19855,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -19870,15 +19870,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -20087,9 +20087,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -20099,7 +20099,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -20112,8 +20112,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -20122,11 +20122,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20137,11 +20137,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20152,15 +20152,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -20357,9 +20357,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -20369,7 +20369,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -20382,8 +20382,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -20392,11 +20392,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20407,11 +20407,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20422,15 +20422,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -20642,9 +20642,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -20654,7 +20654,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -20667,8 +20667,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -20677,11 +20677,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20692,11 +20692,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20707,15 +20707,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -20915,9 +20915,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -20927,7 +20927,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -20940,8 +20940,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -20950,11 +20950,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20965,11 +20965,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -20980,15 +20980,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -21203,9 +21203,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -21215,7 +21215,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -21228,8 +21228,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -21238,11 +21238,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -21253,11 +21253,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -21268,15 +21268,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -21479,9 +21479,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -21491,7 +21491,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -21504,8 +21504,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -21514,11 +21514,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -21529,11 +21529,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -21544,15 +21544,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -21729,9 +21729,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -21741,7 +21741,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -21754,8 +21754,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -21764,11 +21764,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -21779,11 +21779,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -21794,14 +21794,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -21989,9 +21989,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -22001,7 +22001,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -22014,8 +22014,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -22024,11 +22024,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22039,11 +22039,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22054,14 +22054,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -22252,9 +22252,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -22264,7 +22264,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -22277,8 +22277,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -22287,11 +22287,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22302,11 +22302,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22317,14 +22317,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -22503,9 +22503,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -22515,7 +22515,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -22528,8 +22528,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -22538,11 +22538,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22553,11 +22553,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22568,14 +22568,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -22757,9 +22757,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -22769,7 +22769,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -22782,8 +22782,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -22792,11 +22792,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22807,11 +22807,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -22822,14 +22822,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -23014,9 +23014,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -23026,7 +23026,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -23039,8 +23039,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -23049,11 +23049,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23064,11 +23064,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23079,14 +23079,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -23274,9 +23274,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -23286,7 +23286,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -23299,8 +23299,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -23309,11 +23309,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23324,11 +23324,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23339,14 +23339,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -23537,9 +23537,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -23549,7 +23549,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -23562,8 +23562,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -23572,11 +23572,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23587,11 +23587,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23602,14 +23602,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -23794,9 +23794,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -23806,7 +23806,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -23819,8 +23819,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -23829,11 +23829,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23844,11 +23844,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -23859,14 +23859,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -24054,9 +24054,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -24066,7 +24066,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -24079,8 +24079,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -24089,11 +24089,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24104,11 +24104,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24119,14 +24119,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -24317,9 +24317,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -24329,7 +24329,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -24342,8 +24342,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -24352,11 +24352,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24367,11 +24367,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24382,14 +24382,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -24568,9 +24568,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -24580,7 +24580,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -24593,8 +24593,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -24603,11 +24603,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24618,11 +24618,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24633,14 +24633,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -24822,9 +24822,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -24834,7 +24834,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -24847,8 +24847,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -24857,11 +24857,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24872,11 +24872,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -24887,14 +24887,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -25079,9 +25079,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -25091,7 +25091,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -25104,8 +25104,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -25114,11 +25114,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25129,11 +25129,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25144,14 +25144,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -25339,9 +25339,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -25351,7 +25351,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -25364,8 +25364,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -25374,11 +25374,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25389,11 +25389,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25404,14 +25404,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -25602,9 +25602,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -25614,7 +25614,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -25627,8 +25627,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -25637,11 +25637,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25652,11 +25652,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25667,14 +25667,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -25859,9 +25859,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -25871,7 +25871,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -25884,8 +25884,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -25894,11 +25894,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25909,11 +25909,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -25924,14 +25924,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -26119,9 +26119,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -26131,7 +26131,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -26144,8 +26144,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -26154,11 +26154,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26169,11 +26169,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26184,14 +26184,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -26382,9 +26382,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -26394,7 +26394,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -26407,8 +26407,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -26417,11 +26417,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26432,11 +26432,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26447,14 +26447,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -26633,9 +26633,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -26645,7 +26645,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -26658,8 +26658,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -26668,11 +26668,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26683,11 +26683,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26698,14 +26698,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -26887,9 +26887,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -26899,7 +26899,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -26912,8 +26912,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -26922,11 +26922,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26937,11 +26937,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -26952,14 +26952,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -27144,9 +27144,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -27156,7 +27156,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -27169,8 +27169,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -27179,11 +27179,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27194,11 +27194,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27209,14 +27209,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -27404,9 +27404,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -27416,7 +27416,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -27429,8 +27429,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -27439,11 +27439,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27454,11 +27454,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27469,14 +27469,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -27667,9 +27667,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -27679,7 +27679,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -27692,8 +27692,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -27702,11 +27702,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27717,11 +27717,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27732,14 +27732,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -27924,9 +27924,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -27936,7 +27936,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -27949,8 +27949,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -27959,11 +27959,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27974,11 +27974,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -27989,14 +27989,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -28184,9 +28184,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -28196,7 +28196,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -28209,8 +28209,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -28219,11 +28219,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -28234,11 +28234,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -28249,14 +28249,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -28447,9 +28447,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -28459,7 +28459,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -28472,8 +28472,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -28482,11 +28482,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -28497,11 +28497,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -28512,14 +28512,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -28698,9 +28698,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -28710,7 +28710,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -28723,8 +28723,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -28733,11 +28733,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -28748,11 +28748,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -28763,14 +28763,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -28952,9 +28952,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -28964,7 +28964,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -28977,8 +28977,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -28987,11 +28987,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29002,11 +29002,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29017,14 +29017,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -29209,9 +29209,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -29221,7 +29221,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -29234,8 +29234,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -29244,11 +29244,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29259,11 +29259,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29274,14 +29274,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -29469,9 +29469,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -29481,7 +29481,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -29494,8 +29494,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -29504,11 +29504,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29519,11 +29519,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29534,14 +29534,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -29732,9 +29732,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -29744,7 +29744,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -29757,8 +29757,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -29767,11 +29767,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29782,11 +29782,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -29797,14 +29797,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -29989,9 +29989,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -30001,7 +30001,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -30014,8 +30014,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -30024,11 +30024,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30039,11 +30039,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30054,14 +30054,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -30249,9 +30249,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -30261,7 +30261,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -30274,8 +30274,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -30284,11 +30284,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30299,11 +30299,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30314,14 +30314,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -30512,9 +30512,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -30524,7 +30524,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -30537,8 +30537,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -30547,11 +30547,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30562,11 +30562,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30577,14 +30577,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -30763,9 +30763,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -30775,7 +30775,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -30788,8 +30788,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -30798,11 +30798,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30813,11 +30813,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -30828,14 +30828,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -31017,9 +31017,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -31029,7 +31029,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -31042,8 +31042,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -31052,11 +31052,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31067,11 +31067,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31082,14 +31082,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -31274,9 +31274,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -31286,7 +31286,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -31299,8 +31299,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -31309,11 +31309,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31324,11 +31324,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31339,14 +31339,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -31534,9 +31534,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -31546,7 +31546,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -31559,8 +31559,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -31569,11 +31569,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31584,11 +31584,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31599,14 +31599,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -31797,9 +31797,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -31809,7 +31809,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -31822,8 +31822,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -31832,11 +31832,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31847,11 +31847,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -31862,14 +31862,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -32054,9 +32054,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -32066,7 +32066,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -32079,8 +32079,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -32089,11 +32089,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32104,11 +32104,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32119,14 +32119,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -32314,9 +32314,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -32326,7 +32326,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -32339,8 +32339,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -32349,11 +32349,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32364,11 +32364,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32379,14 +32379,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -32577,9 +32577,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -32589,7 +32589,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -32602,8 +32602,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -32612,11 +32612,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32627,11 +32627,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32642,14 +32642,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -32828,9 +32828,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -32840,7 +32840,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -32853,8 +32853,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -32863,11 +32863,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32878,11 +32878,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -32893,14 +32893,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -33082,9 +33082,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -33094,7 +33094,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -33107,8 +33107,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -33117,11 +33117,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33132,11 +33132,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33147,14 +33147,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -33339,9 +33339,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -33351,7 +33351,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -33364,8 +33364,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -33374,11 +33374,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33389,11 +33389,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33404,14 +33404,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -33599,9 +33599,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -33611,7 +33611,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -33624,8 +33624,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -33634,11 +33634,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33649,11 +33649,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33664,14 +33664,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -33862,9 +33862,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -33874,7 +33874,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -33887,8 +33887,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -33897,11 +33897,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33912,11 +33912,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -33927,14 +33927,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -34119,9 +34119,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -34131,7 +34131,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -34144,8 +34144,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -34154,11 +34154,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34169,11 +34169,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34184,14 +34184,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -34379,9 +34379,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -34391,7 +34391,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -34404,8 +34404,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -34414,11 +34414,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34429,11 +34429,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34444,14 +34444,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -34642,9 +34642,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -34654,7 +34654,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -34667,8 +34667,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -34677,11 +34677,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34692,11 +34692,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34707,14 +34707,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -34893,9 +34893,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -34905,7 +34905,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -34918,8 +34918,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -34928,11 +34928,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34943,11 +34943,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -34958,14 +34958,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -35147,9 +35147,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -35159,7 +35159,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -35172,8 +35172,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -35182,11 +35182,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35197,11 +35197,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35212,14 +35212,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -35404,9 +35404,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -35416,7 +35416,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -35429,8 +35429,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -35439,11 +35439,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35454,11 +35454,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35469,14 +35469,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -35664,9 +35664,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -35676,7 +35676,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -35689,8 +35689,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -35699,11 +35699,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35714,11 +35714,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35729,14 +35729,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -35927,9 +35927,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -35939,7 +35939,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -35952,8 +35952,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -35962,11 +35962,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35977,11 +35977,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -35992,14 +35992,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -36184,9 +36184,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -36196,7 +36196,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -36209,8 +36209,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -36219,11 +36219,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -36234,11 +36234,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -36249,14 +36249,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -36444,9 +36444,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -36456,7 +36456,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -36469,8 +36469,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -36479,11 +36479,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -36494,11 +36494,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -36509,14 +36509,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -36707,9 +36707,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -36719,7 +36719,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -36732,8 +36732,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -36742,11 +36742,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -36757,11 +36757,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -36772,14 +36772,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -36958,9 +36958,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -36970,7 +36970,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -36983,8 +36983,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -36993,11 +36993,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37008,11 +37008,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37023,14 +37023,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -37212,9 +37212,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -37224,7 +37224,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -37237,8 +37237,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -37247,11 +37247,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37262,11 +37262,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37277,14 +37277,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -37469,9 +37469,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -37481,7 +37481,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -37494,8 +37494,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -37504,11 +37504,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37519,11 +37519,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37534,14 +37534,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -37729,9 +37729,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -37741,7 +37741,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -37754,8 +37754,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -37764,11 +37764,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37779,11 +37779,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -37794,14 +37794,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -37992,9 +37992,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -38004,7 +38004,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -38017,8 +38017,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -38027,11 +38027,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38042,11 +38042,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38057,14 +38057,14 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+              "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -38241,9 +38241,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -38253,7 +38253,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -38266,8 +38266,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -38276,11 +38276,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38291,11 +38291,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38306,15 +38306,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -38475,9 +38475,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -38487,7 +38487,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -38500,8 +38500,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -38510,11 +38510,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38525,11 +38525,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38540,15 +38540,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+              "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -38718,9 +38718,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -38730,7 +38730,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -38743,8 +38743,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -38753,11 +38753,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.ca.shared.global.16 declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b7c596a91052b75958df78b963338ca733b9b53643eb4b669015b106f926a4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38768,11 +38768,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-all sixteen-byte call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 16.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 99c87c9201c30820920f63624477dfab9046044be0c1e4e9e878a8bd5edd425f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -38783,15 +38783,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d4109cb4c2deaedb98fce7f124afb56882fc7ca5ef239a53d86a293d152b465.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -38970,9 +38970,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -38982,7 +38982,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -38995,8 +38995,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -39005,11 +39005,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.ca.shared.global.4 declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 62367bcc29688f386037306dee01e279908da96b4fbc872a0731f6c1efe3b979.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39020,11 +39020,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-all four-byte call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 4.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fc90fb194ad06cc37e4ec41495d29b2ced1dadd23700f4a6742ba5cb274f236c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39035,15 +39035,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61893d39abe94a66af4946061499f48731945913886e7c08b8e5c3761e2e2089.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -39222,9 +39222,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -39234,7 +39234,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -39247,8 +39247,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -39257,11 +39257,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.ca.shared.global.8 declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b0555ffe6d212958fd651503471d80d0a62030dc35f1dbdac5aedd702be82dd9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39272,11 +39272,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-all eight-byte call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 8.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4115fad6c387b8a02ba11bff217ad8d8f244b96f080aad856ea45b4475ee27f8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39287,15 +39287,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2b95d90d47b38d58ceb6c07c46652eac5c8dec1e1ae2ede4f13972c17c1b268.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -39485,9 +39485,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -39497,7 +39497,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -39510,8 +39510,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -39520,11 +39520,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.ca.shared.global.16.s declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4eb30d0f094641ac8dc5af009c2818cf66555f890dc7e6ed4c8e92987f6ee493.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39535,11 +39535,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-all sixteen-byte source-size calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted both register and immediate source-size forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91e8999a4e6f9db5fd5ba08aa79bd58870a77e05e12b06523f182ec6cfb7a14d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39550,15 +39550,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d5bc6dde1a35b6d38bf05e0d3aea8d44520505a79dedfb9bc4fa3f96543e9c6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -39751,9 +39751,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -39763,7 +39763,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -39776,8 +39776,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -39786,11 +39786,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.ca.shared.global.4.s declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7c8298238217b06520a14baa7dd9017ef6824f9a237ff1c1059c16bcf4a79797.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39801,11 +39801,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-all four-byte source-size calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted both register and immediate source-size forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3022aaec048152a29d81d9b43ad7121669183c87def7ffd6bcc83d52d7040f91.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -39816,15 +39816,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9928045f3f05dc39ca94c224c2afe7c21092310cdfa83a18da46628fa549c1cc.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -40017,9 +40017,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -40029,7 +40029,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -40042,8 +40042,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -40052,11 +40052,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.ca.shared.global.8.s declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0b171b000534f40633027c1b9b7ab26b739a7ebb686ec8617019f3834cad070.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40067,11 +40067,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-all eight-byte source-size calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted both register and immediate source-size forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a01e92a24c61a5a33a96cee208154315907b4538b97345be0e979839c5358c02.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40082,15 +40082,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa2dd28dc4c740dac8fce5c55f0002a8bf7e9ab2210bacfd152777eb0e1928b0.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -40272,9 +40272,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -40284,7 +40284,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -40297,8 +40297,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -40307,11 +40307,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.cg.shared.global.16 declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f32b02e50561865fcbd41af2ab4e6d8c12485b436b14d191cbf5bbf0d248ea82.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40322,11 +40322,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-global sixteen-byte call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.cg.shared.global with copy size 16.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f72afd215027bcf67444625672135ba3fbbfadbee3e28f8e90ced82b0321419e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40337,15 +40337,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e091384fbc0457e593996514ad9eeafc04773c58054a8bc85a8c96980c139e5.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -40535,9 +40535,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -40547,7 +40547,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -40560,8 +40560,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -40570,11 +40570,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.cg.shared.global.16.s declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ce9e52734dd2438c5fc127548069d9cb7130610dc9e0752f6598268db9a3a8fe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40585,11 +40585,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed AS3/AS1 cp.async cache-global sixteen-byte source-size calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted both register and immediate source-size forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5081c35375c4d19aaa528f8abbeb82c57237e9596018318466d212559d1dd65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40600,15 +40600,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a0e1e9bc0608c3d6581f78c2e17d4ef3e71edcd3570f4bde0d9e8f545c8abd6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -40774,9 +40774,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-commit-group"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -40786,7 +40786,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -40799,8 +40799,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -40809,11 +40809,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.commit.group declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the zero-operand declaration and canonicalized it as nounwind.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ea4574621afbb04281bfa8ab26db49b9a256ea5a991c63ff2907c9a725aef5a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40824,11 +40824,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed zero-operand cp.async commit-group call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.commit_group.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b2f1831e64e9e04bbe61a19f62e926eebb9e8c7a2e6e7f3fc073d97363c81f2a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -40839,15 +40839,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 533d9b5b9a9c927d1fe0b23c76106771cdac35407624486b921bcf4c7096811a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -41004,9 +41004,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -41016,7 +41016,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -41029,8 +41029,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -41040,11 +41040,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive(ptr) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the generic-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3d6f3f642d9e80245cb2f6c76442ccd0d824326e288898d121d5645368edcc8e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41055,11 +41055,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed generic-address cp.async mbarrier arrival",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8b4dfc81ee704ec2a8f8e8502b5f5673cafe35607eb28120b57fff0c3b27baf0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41070,15 +41070,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM bridge probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 86913ed12af6b0bd59e448e3b5ccc6823e2f0b6a140146ead578815ea6f7c5ec.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -41272,9 +41272,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -41284,7 +41284,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -41297,8 +41297,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -41308,11 +41308,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive.noinc(ptr) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the generic-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11cd942627957bdbe07521415b591132bac6543a6d37a10286821ca4be498129.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41323,11 +41323,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed generic-address no-increment cp.async mbarrier arrival",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.noinc.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d65ebc7c63cdfa57f0f5aefa206fd07f6004879beb44350640071d5eb97e6625.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41338,15 +41338,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM bridge probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ef8193aee88f5efa25293d77c1399be0b83dd294401561177e91f144714072a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -41541,9 +41541,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -41553,7 +41553,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -41566,8 +41566,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -41577,11 +41577,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive.noinc.shared(ptr addrspace(3)) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b62ad0b051d6ccaa435c3686d1a94547c3635490996a003e78abb772b160cc4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41592,11 +41592,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed shared-address no-increment cp.async mbarrier arrival",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.noinc.shared.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9affe7627501776e21eb8059bdb4bee726a956435720cc6de05ee4d93ef51be1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41607,15 +41607,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM bridge probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 366217f18f9473cf5ccc1d87632a590d8fcf7a8f5508a5bf876c49df1dc939e2.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -41811,9 +41811,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -41823,7 +41823,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -41836,8 +41836,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -41847,11 +41847,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive.shared(ptr addrspace(3)) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cd2056552d6e2b796c195f31ad768ca175eac447825d819f5957b9fcec07059c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41862,11 +41862,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed shared-address cp.async mbarrier arrival",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.shared.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d59be5c8527b799671b85f003a16a51bbd07448cc2130863d25e7318f2841384.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -41877,15 +41877,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM bridge probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 23bdc05d4be1c27246a496b4f2ebede69efa8b84d326395adc4d804f16f9eef5.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -42071,9 +42071,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -42083,7 +42083,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -42096,8 +42096,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -42106,11 +42106,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.wait.all declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the zero-operand declaration and canonicalized it as nounwind.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1b3dc9f7a7dd7ebd28ad446ba1b8a97a30ffa6065ab3b9bdbb2bcb4377287859.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42121,11 +42121,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed zero-operand cp.async wait-all call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.wait_all.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2dfc842c70441fe57d5a892d34441cf37fedab3955f511c5ef23118632e06924.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42136,15 +42136,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 96ab3b4af8cae0f8d415560129805aec75b9e3e8c26b916cda16a42cfc11d581.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -42300,9 +42300,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -42312,7 +42312,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -42325,8 +42325,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -42335,11 +42335,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed llvm.nvvm.cp.async.wait.group(i32 immarg) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and canonicalized its operand as immarg and the function as nounwind.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c4b4d445968d5e8db78f64d6903e5dab71d7e0a7ed2edd66bff7dd5443900376.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42350,11 +42350,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed constant cp.async wait-group call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact cp.async.wait_group with an integer literal.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0604968fa8260d2a46de521f458b9ea4da639127e0671340e531eb5af1e5bea6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42365,15 +42365,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM cp.async probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 77ca31b604849a0fbb4a6eeff8a57d146053d576721be6dc2212ec7347f0f1c1.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -42533,9 +42533,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -42545,7 +42545,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -42558,8 +42558,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -42568,11 +42568,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed intrinsic llvm.nvvm.ff2f16x2.rn(float, float) -> <2 x half>",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 23c48425228bb21090b442cbcb63b58874efaf6cc05dde2ad7549aff23e3f8ce.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42583,11 +42583,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed call to llvm.nvvm.ff2f16x2.rn(float high, float low) -> <2 x half>",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7c94821ab335a6f02f2b3ca52d1e0c2428d1d26aac5e7c62852ef35cbfcba8e4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42598,26 +42598,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX cvt.rn.f16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rn, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 009a800094f29371a5f120862eb39d45b097aaa5715ce2e7ac26a43ec3665773.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42628,15 +42628,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -42805,9 +42805,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -42817,7 +42817,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -42830,8 +42830,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -42840,11 +42840,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed intrinsic llvm.nvvm.ff2bf16x2.rn(float, float) -> <2 x bfloat>",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7b66943f3e72c942117544102ede74b7d46537dc9c8f09cb4c800b1a8150b25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42855,11 +42855,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed call to llvm.nvvm.ff2bf16x2.rn(float high, float low) -> <2 x bfloat>",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 260159c4d2bde9577b33387b20e93e5e1e00605c7ce69ce157c5c5046bbaabf0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42870,26 +42870,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX cvt.rn.bf16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rn, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 625e5da377d73334b52ec4bdd6bc6702e15b196ca54350f0b070cb61e09b60ff.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -42900,15 +42900,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -43074,9 +43074,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -43086,7 +43086,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -43099,8 +43099,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -43110,11 +43110,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b295a2577d8a238ef3098f7cb0e4fc309f9e5f1177f9f02ae766101f9d70cc67.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43125,11 +43125,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdceb080a053c25340804b87375d5fc8693a5fc8eaa461399d438982c1cd179a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43140,26 +43140,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5682cacdfec0f3e7fe02db10b1665b3b88ab176b4db26738c965155f31ba0dbf.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43170,15 +43170,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -43356,9 +43356,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -43368,7 +43368,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -43381,8 +43381,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -43392,11 +43392,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a572f8ac9426c4acabd244748fbcabfffc19a7afbc3eb6b74b087a7397d0ca5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43407,11 +43407,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 182d55254c86ae42a4eaaf4db2c9829f771eb0f23f6abaa92f773fd9910b607b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43422,26 +43422,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7d96da22340431d9ae279f1ea4d4573bca35a171e7fc6199b5893b540a93424.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43452,15 +43452,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -43641,9 +43641,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -43653,7 +43653,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -43666,8 +43666,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -43676,11 +43676,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed intrinsic llvm.nvvm.ff2bf16x2.rn.relu(float, float) -> <2 x bfloat>",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e02262263ec043530dec4469b22cfd6e831bc597d97b59e20132167985ae8e75.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43691,11 +43691,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed call to llvm.nvvm.ff2bf16x2.rn.relu(float high, float low) -> <2 x bfloat>",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.relu.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c797710407ad5b5c6425fc3ae32587456573e2442fb1398a784e61ab025c7f6b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43706,26 +43706,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX cvt.rn.relu.bf16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.relu.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rn.relu, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 336ac20640c8b6451d754f6afed9479d89a22cd5c47b6606e57f251337b35e0c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43736,15 +43736,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -43911,9 +43911,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -43923,7 +43923,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -43936,8 +43936,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -43947,11 +43947,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e12dd4763bbc1533bd40687021b8319bd843af8e62fc1fefe8212335310449ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43962,11 +43962,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e0264fce6d34633b3225bd3f48723a6efaa11b18a738168e34c460dfe2b16754.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -43977,26 +43977,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a2341dac9ffd42fede2aefd66f576cac3e464c0b6b81c38773128832579eafa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44007,15 +44007,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -44194,9 +44194,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -44206,7 +44206,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -44219,8 +44219,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -44230,11 +44230,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b32d14ca011f505a526f0fecf85660bd93e1bbc147e728896df311e7f2b1176b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44245,11 +44245,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f69dc7503144fa82436f92c37e77254ebf11ca61ffdcf492ce8b6ace6ae00234.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44260,26 +44260,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dff12609bfb2cbd5258c5d20e2e1c8b781090475d157dfe239cdbfc5edd9c204.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44290,15 +44290,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -44480,9 +44480,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -44492,7 +44492,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -44505,8 +44505,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -44515,11 +44515,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed intrinsic llvm.nvvm.ff2f16x2.rn.relu(float, float) -> <2 x half>",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f4b2d11b6f47357e52b5da2bd8374d6043945e2d3c7d28f0d9bce741658cbbfb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44530,11 +44530,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed call to llvm.nvvm.ff2f16x2.rn.relu(float high, float low) -> <2 x half>",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.relu.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a886266d7776914c367d748d72bd9b1032db99c5eb5366f902a3b5b8e63f2add.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44545,26 +44545,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX cvt.rn.relu.f16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.relu.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rn.relu, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 644949d91a98e2c1b36b125ca7211d181b486d46e6b0335c823f0f78de9860cc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44575,15 +44575,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -44759,9 +44759,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -44771,7 +44771,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -44784,8 +44784,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -44795,11 +44795,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "canonical typed LLVM declarations for the rn/rz satfinite and relu.satfinite conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44810,11 +44810,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM calls for the four PTX 8.6 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -44825,15 +44825,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -45024,9 +45024,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -45036,7 +45036,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -45049,8 +45049,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -45060,11 +45060,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "canonical typed LLVM declarations for the rn, rn.relu, rz, and rz.relu conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45075,11 +45075,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM calls for the four PTX 7.8 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45090,15 +45090,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -45279,9 +45279,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -45291,7 +45291,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -45304,8 +45304,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -45315,11 +45315,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 777bd7a5362a935cb9b45c99eb5c7912d8296fd322f232be5feb4dcb71afe958.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45330,11 +45330,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8843a1e5b9fee11892601ac9688d35a5195bdd8c88122bb6ab5b4fa21cf76e29.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45345,26 +45345,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d344820320e9db759a6db9eebd370929159ce71d483a76cb70f846d27abf426.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45375,15 +45375,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -45565,9 +45565,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -45577,7 +45577,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -45590,8 +45590,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -45601,11 +45601,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all four packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45616,11 +45616,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls in PTX high, low source order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45631,26 +45631,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with u16 output and public low, high inputs",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45661,15 +45661,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -45851,9 +45851,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -45863,7 +45863,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -45876,8 +45876,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -45887,11 +45887,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b6fcb8ea52dc533221fc3a953835ab9fb2cdb8a3c82a1dc9527dc2b980cb096.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45902,11 +45902,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6707ccb77a07e8b004032798bbfd07afba3af30d15568ac2c3d08d84d8b10b3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45917,26 +45917,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6064f89491e6f84f9ac56c6373eec145a9b8d55507b30a4ee2ffcbe2deed7dd1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -45947,15 +45947,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -46137,9 +46137,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -46149,7 +46149,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -46162,8 +46162,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -46173,11 +46173,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all four packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46188,11 +46188,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls in PTX high, low source order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46203,26 +46203,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with u16 output and public low, high inputs",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46233,15 +46233,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -46423,9 +46423,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -46435,7 +46435,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -46448,8 +46448,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -46459,11 +46459,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a17f8309a9b1436ceae14d34278cbc25993c1607850ec87962d7fef9fd6c3215.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46474,11 +46474,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11284b3464e72006408b832095a07a2ee8683336b367d675109a20b5289db239.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46489,26 +46489,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 82c1a28789b05b8df6d92cc52932afdc99dc761cb2bcc77c7d85e2b3a1da1e24.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46519,15 +46519,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -46710,9 +46710,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -46722,7 +46722,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -46735,8 +46735,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -46746,11 +46746,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all four packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46761,11 +46761,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls in PTX high, low source order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46776,26 +46776,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with u16 output and public low, high inputs",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -46806,15 +46806,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -46997,9 +46997,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -47009,7 +47009,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -47022,8 +47022,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -47033,11 +47033,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2a8ce890aff0fa2df126aba76cf0dcc3fbda114de759ad215043e6ad0d400d23.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47048,11 +47048,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls on a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab076ae9d2de65d0ab3caad272a118bd426726141914f2b539a30a616d655e4e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47063,26 +47063,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with a single packed source operand",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cf21d0dd82dfc3a278f16fc60cf6772892bc2faee051ae9dd92b23f6e934f314.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47093,15 +47093,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -47284,9 +47284,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -47296,7 +47296,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -47309,8 +47309,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -47320,11 +47320,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "canonical LLVM declarations for all four packed-FP8 conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47335,11 +47335,11 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM intrinsic calls in PTX high, low source order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47350,26 +47350,26 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "pure inline PTX with u16 output and public low, high inputs",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47380,15 +47380,15 @@
                 "sm_89",
                 "ptx81"
               ],
-              "representation": "inline PTX assembled at the exact floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -47580,9 +47580,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -47592,7 +47592,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -47605,8 +47605,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -47616,11 +47616,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "canonical typed LLVM declarations for the rn/rz satfinite and relu.satfinite conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47631,11 +47631,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM calls for the four PTX 8.6 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47646,15 +47646,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -47844,9 +47844,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -47856,7 +47856,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -47869,8 +47869,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -47880,11 +47880,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "canonical typed LLVM declarations for the rn, rn.relu, rz, and rz.relu conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47895,11 +47895,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM calls for the four PTX 7.8 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -47910,15 +47910,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -48107,9 +48107,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -48119,7 +48119,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.1",
             "hardware": {
@@ -48132,8 +48132,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -48143,11 +48143,11 @@
                 "sm_80",
                 "ptx81"
               ],
-              "representation": "canonical typed LLVM declaration for cvt.rna.satfinite.tf32.f32",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the fresh 628-byte probe to 634-byte IR (SHA-256 51e877e213ff97519d0a1c4080afee97c3e4b208b88ab2071b214a36428c77f4) with the imported f32-to-i32 signature and canonical intrinsic attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a38ab8cd30065c8554fc11f7b6f155740eb533d290525b57a45a4b339465234.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48158,11 +48158,11 @@
                 "sm_80",
                 "ptx81"
               ],
-              "representation": "typed llvm.nvvm.f2tf32.rna.satfinite call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc lowered the fresh probe to 634-byte PTX (SHA-256 be00dc0e407a9f71d14c5b73a5a4c53b06e4430ad20f888e5595499b25c348ad) containing cvt.rna.satfinite.tf32.f32.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2819c3052033ee04b85149f6891347ea860e7bacc511a0e7edb1d5fb4f36cd2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48173,15 +48173,15 @@
                 "sm_80",
                 "ptx81"
               ],
-              "representation": "typed LLVM output assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the fresh sm_80/PTX 8.1 output and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e8c2a92a43cdb367ea7856ca832f6fea44f45613b70f89f2acd2b45c4d8b35b4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -48371,9 +48371,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -48383,7 +48383,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -48396,8 +48396,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -48407,11 +48407,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "canonical typed LLVM declaration for cvt.rna.tf32.f32",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the fresh 598-byte probe to 584-byte IR (SHA-256 2448daf9235fa4db9287c592309f7496856e9cf31929014a569c70813978d9c8) with the imported f32-to-i32 signature and canonical intrinsic attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7fc68746b0ee106bcb3270f8a02cb83cd0a4e85363c0854f82546717a33bf72d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48422,11 +48422,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.f2tf32.rna call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc lowered the fresh probe to 565-byte PTX (SHA-256 19695568a4fde5c5847f25e5103f5f4aecf6b96dd9f408028fee66222fe8c82f) containing cvt.rna.tf32.f32.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5355197dbd45ff46716deac34ba4d8de1084ee4374f537af9a67dd2fb47b6c5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48437,15 +48437,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM output assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the fresh sm_80/PTX 7.0 output and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8bf2670d9bea4e04e196ffe05f2954a5d9f8beefc57a231af094444b20348c3a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -48628,9 +48628,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -48640,7 +48640,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -48653,8 +48653,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -48663,11 +48663,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed intrinsic llvm.nvvm.ff2bf16x2.rz(float, float) -> <2 x bfloat>",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb6bcfe8642c6981459470ff9b577231bb7c5575846df5a6b524e3ed834bfc4c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48678,11 +48678,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed call to llvm.nvvm.ff2bf16x2.rz(float high, float low) -> <2 x bfloat>",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rz.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dcafda6629d6de7301da44bf1e7690a5d1e8131d595cc975ea4e8b68a61de5d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48693,26 +48693,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX cvt.rz.bf16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rz.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rz, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd97b8bfbf98ce29b538e25417e266f1d5c29d1d18b0789af1dd439fe4431d90.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48723,15 +48723,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -48900,9 +48900,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -48912,7 +48912,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -48925,8 +48925,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -48935,11 +48935,11 @@
               "targets": [
                 "sm_80"
               ],
-              "representation": "typed intrinsic llvm.nvvm.ff2f16x2.rz(float, float) -> <2 x half>",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2bccf7085ce3ccc76e323145e0cd1e86fda391d0411adc51c1eea4ce806a2f03.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48950,11 +48950,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed call to llvm.nvvm.ff2f16x2.rz(float high, float low) -> <2 x half>",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rz.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5790119795c6ad9cbfc37ff239e9a123b44c05bde5d38fc5675565933d23a6ff.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48965,26 +48965,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX cvt.rz.f16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rz.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rz, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6833e6724886aa52db6de2c812bc6e0e2a7216e33ba07501701559eb4c68cf5c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -48995,15 +48995,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -49178,9 +49178,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -49190,7 +49190,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -49203,8 +49203,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -49214,11 +49214,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "canonical typed LLVM declarations for the rn/rz satfinite and relu.satfinite conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -49229,11 +49229,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM calls for the four PTX 8.6 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -49244,15 +49244,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -49443,9 +49443,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -49455,7 +49455,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -49468,8 +49468,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -49479,11 +49479,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "canonical typed LLVM declarations for the rn, rn.relu, rz, and rz.relu conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -49494,11 +49494,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM calls for the four PTX 7.8 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -49509,15 +49509,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -49707,9 +49707,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -49719,7 +49719,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -49732,8 +49732,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -49743,11 +49743,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "canonical typed LLVM declarations for the rn/rz satfinite and relu.satfinite conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -49758,11 +49758,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM calls for the four PTX 8.6 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -49773,15 +49773,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -49971,9 +49971,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -49983,7 +49983,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -49996,8 +49996,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -50007,11 +50007,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "canonical typed LLVM declarations for the rn, rn.relu, rz, and rz.relu conversions",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50022,11 +50022,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM calls for the four PTX 7.8 scalar conversions",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50037,15 +50037,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed LLVM outputs assembled at the catalog floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -50233,9 +50233,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -50245,7 +50245,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -50258,8 +50258,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -50269,11 +50269,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50284,11 +50284,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50299,15 +50299,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -50496,9 +50496,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -50508,7 +50508,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -50521,8 +50521,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -50532,11 +50532,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50547,11 +50547,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50562,15 +50562,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -50759,9 +50759,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -50771,7 +50771,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -50784,8 +50784,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -50795,11 +50795,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50810,11 +50810,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -50825,15 +50825,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -51023,9 +51023,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -51035,7 +51035,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -51048,8 +51048,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -51059,11 +51059,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51074,11 +51074,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51089,15 +51089,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -51286,9 +51286,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -51298,7 +51298,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -51311,8 +51311,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -51322,11 +51322,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51337,11 +51337,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51352,15 +51352,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -51549,9 +51549,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -51561,7 +51561,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -51574,8 +51574,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -51585,11 +51585,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51600,11 +51600,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51615,15 +51615,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -51813,9 +51813,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -51825,7 +51825,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -51838,8 +51838,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -51849,11 +51849,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51864,11 +51864,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -51879,15 +51879,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -52076,9 +52076,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -52088,7 +52088,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -52101,8 +52101,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -52112,11 +52112,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52127,11 +52127,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52142,15 +52142,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -52339,9 +52339,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -52351,7 +52351,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -52364,8 +52364,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -52375,11 +52375,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52390,11 +52390,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52405,15 +52405,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -52603,9 +52603,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -52615,7 +52615,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -52628,8 +52628,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -52639,11 +52639,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52654,11 +52654,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52669,15 +52669,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -52866,9 +52866,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -52878,7 +52878,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -52891,8 +52891,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -52902,11 +52902,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52917,11 +52917,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -52932,15 +52932,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -53129,9 +53129,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -53141,7 +53141,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -53154,8 +53154,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -53165,11 +53165,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53180,11 +53180,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53195,15 +53195,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -53409,9 +53409,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_61",
@@ -53421,7 +53421,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "5.0",
             "hardware": {
@@ -53434,8 +53434,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -53444,11 +53444,11 @@
               "targets": [
                 "sm_61"
               ],
-              "representation": "typed intrinsic llvm.nvvm.idp2a.s.s(i32, i32, i1 immarg, i32)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration, immediate selector, and no-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cf14ce2498510d2ccdf8e26062d9dab33a2f018ab44d7402f731ee721edd9586.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53459,11 +53459,11 @@
                 "sm_61",
                 "ptx50"
               ],
-              "representation": "typed call to llvm.nvvm.idp2a.s.s with selector false",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend mapped false to exact dp2a.lo.s32.s32 PTX at the native floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5004901d9a7fa9fc719218c051ffd4362bcef5a21c7bcca397e74fdb199f3733.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53474,15 +53474,15 @@
                 "sm_75",
                 "ptx50"
               ],
-              "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c5f55a84bfd467a639f8fe3020d3fa05a6616160125d05ecd44cf631c78a954c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -53673,9 +53673,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_61",
@@ -53685,7 +53685,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "5.0",
             "hardware": {
@@ -53698,8 +53698,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -53708,11 +53708,11 @@
               "targets": [
                 "sm_61"
               ],
-              "representation": "typed intrinsic llvm.nvvm.idp2a.u.u(i32, i32, i1 immarg, i32)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration, immediate selector, and no-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6bba537191d91e4c7f577889e7832f879421a5a8b021a98982af2c068ecd5ac5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53723,11 +53723,11 @@
                 "sm_61",
                 "ptx50"
               ],
-              "representation": "typed call to llvm.nvvm.idp2a.u.u with selector false",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend mapped false to exact dp2a.lo.u32.u32 PTX at the native floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ad295ebd24c5fa94a1990f0e676049b7232bf0421f6419e2230f054510b87ede.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53738,15 +53738,15 @@
                 "sm_75",
                 "ptx50"
               ],
-              "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aef5395fd4ebfc37a0a4fb468bb18c83a363528cd307bfeb7ed1218c5fd84fc2. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -53927,9 +53927,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_61",
@@ -53939,7 +53939,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "5.0",
             "hardware": {
@@ -53952,8 +53952,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -53962,11 +53962,11 @@
               "targets": [
                 "sm_61"
               ],
-              "representation": "typed intrinsic llvm.nvvm.idp4a.s.s(i32, i32, i32)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and added its no-memory and speculatable attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 de5ba8b2a4efbc2ca94148badabcdf0d50913a97a1663fb623592433f484e88e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53977,11 +53977,11 @@
                 "sm_61",
                 "ptx50"
               ],
-              "representation": "typed call to llvm.nvvm.idp4a.s.s",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact dp4a.s32.s32 PTX at the native floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9c42148d5c5a275ca9ece93315e98eb0e9508a0b4e8cf0eba8cc761268f29c42.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -53992,15 +53992,15 @@
                 "sm_75",
                 "ptx50"
               ],
-              "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d9b7b74a68441f7f40d480ab430166c946704ea11caf1a7e9951b4726130bc65. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -54180,9 +54180,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_61",
@@ -54192,7 +54192,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "5.0",
             "hardware": {
@@ -54205,8 +54205,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -54215,11 +54215,11 @@
               "targets": [
                 "sm_61"
               ],
-              "representation": "typed intrinsic llvm.nvvm.idp4a.u.u(i32, i32, i32)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and added its no-memory and speculatable attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4565681a3a6f6f64ea89b63e96c44c1aef2d594b01f3c695eaa4d35147c250f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -54230,11 +54230,11 @@
                 "sm_61",
                 "ptx50"
               ],
-              "representation": "typed call to llvm.nvvm.idp4a.u.u",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact dp4a.u32.u32 PTX at the native floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eec368e7d9686a4072001c8362bc69ea91a9b25fb950b50232244c96e72c0887.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -54245,15 +54245,15 @@
                 "sm_75",
                 "ptx50"
               ],
-              "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 159beb8496be094d32d2c6421e8993b91ee8e0d8b787fa5313af73ea5af66eac. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -54403,9 +54403,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-ld"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -54415,7 +54415,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-cluster-memory",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-cluster-memory",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -54428,8 +54428,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -54439,11 +54439,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "one convergent LLVM kernel containing exact mapa and remote u32-load inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 690-byte source had SHA-256 ad8a94dfa560d523bfe8cf963f8779132c960e1e6e5717e080e72819936f446a; the 1,007-byte PTX had SHA-256 005a7f90c8a3bd97a99792ce8a181d43c28fb318cd20dce6068fcb349800a951.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418a46a6a0279c0e2c7391fcccae3b3bec1a660a3ffad28b1df2a7f63df297d6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -54454,15 +54454,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX retaining both cluster-memory operations",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,712-byte cubin with SHA-256 1780aee43cd691bd7e3a79f4bc4b2d163e5cdfc11e08441ae318dfb091a5057a.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bf4f5d63cf41eae708cc3087c76cbea8b9812ba26f1c915646ef4b0205c51760.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -54623,9 +54623,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-dynamic-smem-size"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -54635,7 +54635,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "4.1",
             "hardware": {
@@ -54648,8 +54648,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -54659,11 +54659,11 @@
                 "sm_20",
                 "ptx41"
               ],
-              "representation": "pure inline PTX reads of dynamic and total shared-memory sizes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted both size reads in 655-byte PTX with SHA-256 4ab2630ee33912457ed45c1cd516ae0797e062fcba7123db9bac352c67280b3a.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24cc1707db68cd33a711956ac50c8c65ab3e84eb5cf71a00fc15160d770b645d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -54674,15 +54674,15 @@
                 "sm_75",
                 "ptx41"
               ],
-              "representation": "LLVM-produced PTX retaining both shared-memory size reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 ee3358b59d2b596f1b276d983c3149d58c384872ee5edad482cf20a99183e7b1.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7feb7ce1442470cb7ca3f613338c9fa901e16b23bc621a84ac3aa46a8583088d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx41, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -54862,9 +54862,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-elect-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -54874,7 +54874,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-elect",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-elect",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -54887,8 +54887,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -54897,11 +54897,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed llvm.nvvm.elect.sync declaration and register/immediate calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the 1,080-byte source with SHA-256 e83b2cd7839b679e3d23da509c817d79108c84ce4eabefc70282dd2065cf69a1 and produced 2,100-byte bitcode with SHA-256 77043e86454445b633d31ec577a6c27c5a2675b960421a0739a621de3896a9e9.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d6be8db0355a531bcc20af99d6b2ab9a8015c92ce61965522dec78614fd03d1c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -54912,11 +54912,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "typed LLVM probe retaining register and immediate member-mask forms",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM 22 emitted exactly elect.sync %r2|%p1, %r1 and elect.sync %r3|%p2, -1 with no extern calls; the 826-byte PTX had SHA-256 86e7fcebf74be1752a172688c87ec3bf042a296f7846df731b0385ad9145bf23.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 de0e95f3672e5c96d15e4fd737d9225940692889c495211037f7b4f99d3cddb4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -54927,15 +54927,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX retaining both elect.sync selections",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 4,328-byte cubin with SHA-256 f13f6d2c6aa0971a28ecf01f5008727bc513b0acef9cdd075973f03f30c15876.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 192c7ae696bf472409f022926d3c74f98784499ab59d769d1565e6605976c7d0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -55107,9 +55107,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-envreg"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -55119,7 +55119,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -55132,8 +55132,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -55142,11 +55142,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -55157,11 +55157,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed calls lowering to six exact special-register reads",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -55172,15 +55172,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -55344,9 +55344,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-envreg"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -55356,7 +55356,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -55369,8 +55369,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -55379,11 +55379,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -55394,11 +55394,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed calls lowering to six exact special-register reads",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -55409,15 +55409,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -55592,9 +55592,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-ex2"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -55604,7 +55604,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -55617,8 +55617,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -55628,11 +55628,11 @@
                 "sm_75",
                 "ptx70"
               ],
-              "representation": "llvm.nvvm.ex2.approx.f16 lowered at the documented hardware and PTX floor",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc selected ex2.approx.f16 for sm_75 with +ptx70; PTX SHA-256 21a1d2fe684434a6c71b2454ff0a26463002def6ab7c09e00252a70d9770b7df. The same probe fails selection at sm_70 or +ptx65. llvm.nvvm.ex2.approx.ftz.f16 and llvm.nvvm.ex2.approx.bf16 fail selection; llvm.nvvm.ex2.approx.ftz.bf16 selects at sm_90 with +ptx78.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cee684a507daecda40b46b40a8de9a2f2582dfcf5aae88b3d844b49e614c44e7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -55643,15 +55643,15 @@
                 "sm_75",
                 "ptx70"
               ],
-              "representation": "single ex2.approx.f16 instruction assembled at its admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas assembled the 640-byte sm_75/PTX 7.0 output into a 1,464-byte cubin with SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0fd052b8dd9f1821ac995eba3175320422af7c36dcdf0fc99190a1b2819de6e6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -55827,9 +55827,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-ex2"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -55839,7 +55839,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -55852,8 +55852,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -55863,30 +55863,30 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "3 appended inline-PTX unary math probes (ex2.approx.{f32,ftz.f32}, tanh.approx.f32) attested individually",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "3 appended inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -56062,9 +56062,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-ex2"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -56074,7 +56074,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -56087,8 +56087,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -56098,30 +56098,30 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "3 appended inline-PTX unary math probes (ex2.approx.{f32,ftz.f32}, tanh.approx.f32) attested individually",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "3 appended inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -56296,9 +56296,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-mbarrier-init"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -56308,7 +56308,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -56321,8 +56321,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -56332,11 +56332,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -56347,15 +56347,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX 8.0 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -56507,9 +56507,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -56519,7 +56519,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -56532,8 +56532,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -56543,11 +56543,11 @@
                 "sm_90",
                 "ptx86"
               ],
-              "representation": "generated cluster-scope inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -56558,15 +56558,15 @@
                 "sm_90",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX 8.6 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -56720,9 +56720,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -56732,7 +56732,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -56745,8 +56745,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -56756,11 +56756,11 @@
                 "sm_90",
                 "ptx86"
               ],
-              "representation": "generated cluster-scope inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -56771,15 +56771,15 @@
                 "sm_90",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX 8.6 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -56933,9 +56933,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -56945,7 +56945,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -56958,8 +56958,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -56969,11 +56969,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -56984,15 +56984,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX 8.0 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -57154,9 +57154,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -57166,7 +57166,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -57179,8 +57179,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -57189,11 +57189,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57204,11 +57204,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57219,15 +57219,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -57397,9 +57397,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -57409,7 +57409,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -57422,8 +57422,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -57432,11 +57432,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57447,11 +57447,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57462,15 +57462,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -57640,9 +57640,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -57652,7 +57652,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -57665,8 +57665,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -57675,11 +57675,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57690,11 +57690,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57705,15 +57705,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -57883,9 +57883,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -57895,7 +57895,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -57908,8 +57908,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -57918,11 +57918,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57933,11 +57933,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -57948,15 +57948,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -58116,9 +58116,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -58128,7 +58128,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -58141,8 +58141,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -58151,11 +58151,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58166,11 +58166,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58181,15 +58181,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -58341,9 +58341,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -58353,7 +58353,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -58366,8 +58366,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -58376,11 +58376,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58391,11 +58391,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58406,15 +58406,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -58566,9 +58566,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -58578,7 +58578,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -58591,8 +58591,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -58601,11 +58601,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58616,11 +58616,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58631,15 +58631,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -58791,9 +58791,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -58803,7 +58803,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -58816,8 +58816,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -58826,11 +58826,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "eight individually generated typed tensor-map fence declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58841,11 +58841,11 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight typed tensor-map proxy-fence probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -58856,15 +58856,15 @@
                 "sm_90",
                 "ptx83"
               ],
-              "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -59034,9 +59034,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -59046,7 +59046,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -59059,8 +59059,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -59070,11 +59070,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX fma.rn.bf16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact fma.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f847702ab3d3db2051359e9ba8869735a57a5cf5b92a2901b154340dc14cc080.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -59085,15 +59085,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa10765bb36d761080b94f2c0788a77e387b12032c06504740e5ff756239cda5.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -59275,9 +59275,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -59287,7 +59287,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -59300,8 +59300,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -59311,11 +59311,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX fma.rn.f16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact fma.rn.f16x2 inline PTX in the combined four-form PTX 6.0/sm_70 probe. An independent typed call to llvm.nvvm.fma.rn.f16x2 emitted the same instruction at PTX 6.0/sm_70 and at its native PTX 4.2/sm_53 floor; sm_52 failed exact intrinsic selection.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e1cf777d5d76a7579b7c840ac1250cd7961888126aeef92b619e2bfcf5718aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -59326,15 +59326,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f962661dab009881f23af0cdf33aed0bdb5ea13629f790897da04f271d8cc305. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -59487,9 +59487,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -59499,7 +59499,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -59512,8 +59512,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -59523,11 +59523,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -59538,15 +59538,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -59728,9 +59728,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -59740,7 +59740,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -59753,8 +59753,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -59764,11 +59764,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX fma.rn.ftz.f16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80a7eb25c335cf849bc1e0be9ed1e391b9d311a1573d38be927d6aa35c8b4aa3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -59779,15 +59779,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83de32412117576acec7391b12d4c1e5f2f026f4fed9281689ec4a87e9b1deaf. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -59956,9 +59956,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -59968,7 +59968,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -59981,8 +59981,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -59992,11 +59992,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -60007,15 +60007,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -60198,9 +60198,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -60210,7 +60210,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -60223,8 +60223,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -60234,11 +60234,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX fma.rn.ftz.relu.f16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 1 exact instruction head(s) in 682-byte PTX (SHA-256 ee283e72de7ddec2258b4eb628ca180889b67d9f79d13ba926719724bec1c01f) lowered from the 196-byte fixture (SHA-256 fa242c0f862cc57936c53949b30d4ce59871e59cde0349570a44122c3a164006). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93c9c1a8aaa9ac2bee924f4b1eee264e0aa38c8c6f01c04bddf3bfd58d169bc4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -60249,15 +60249,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module at sm_80 and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 582782aeec6a6286313dd28330e3de2a23eeaa07c479ee6271c831e0c8f9d1e7.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -60456,9 +60456,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -60468,7 +60468,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -60481,8 +60481,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -60492,11 +60492,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX fma.rn.ftz.sat.f16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9bed831d23f91ca28a7fa178103ea679a4765ec587f918986874e52b21d00da1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -60507,15 +60507,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6a185d53a7cf04592fc205ee4ef0926f252fa57f5c50f1ee510c7aa6210578ee. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -60714,9 +60714,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -60726,7 +60726,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -60739,8 +60739,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -60750,11 +60750,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX fma.rn.relu.bf16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact fma.rn.relu.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 903e4771d5a308e61a37282dd1f236481315e6519d5c06dbfd19c6c47f97acfd.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -60765,15 +60765,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 adc9face773753523147665457351aa23a8b43da9e697a565109e15870383010.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -60956,9 +60956,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -60968,7 +60968,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -60981,8 +60981,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -60992,11 +60992,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX fma.rn.relu.f16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67da91ea322e2d60af10838e8a20890d99c38e07262abe9a6848ea5e1fbbf2fb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61007,15 +61007,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined three-form sm_80 packed-f16x2 upper probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 871b4ac65868041bcc0f028deaaac2793f5df9db4518fdbfaa03580e9c26414a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -61195,9 +61195,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -61207,7 +61207,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -61220,8 +61220,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -61231,11 +61231,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61246,11 +61246,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61261,15 +61261,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -61465,9 +61465,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -61477,7 +61477,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -61490,8 +61490,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -61501,11 +61501,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61516,11 +61516,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61531,15 +61531,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -61735,9 +61735,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -61747,7 +61747,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -61760,8 +61760,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -61771,11 +61771,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61786,11 +61786,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -61801,15 +61801,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -62006,9 +62006,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -62018,7 +62018,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -62031,8 +62031,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -62042,11 +62042,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62057,11 +62057,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62072,15 +62072,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -62278,9 +62278,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -62290,7 +62290,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -62303,8 +62303,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -62314,11 +62314,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62329,11 +62329,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62344,15 +62344,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -62549,9 +62549,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -62561,7 +62561,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -62574,8 +62574,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -62585,11 +62585,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62600,11 +62600,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62615,15 +62615,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -62819,9 +62819,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -62831,7 +62831,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -62844,8 +62844,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -62855,11 +62855,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62870,11 +62870,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -62885,15 +62885,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -63089,9 +63089,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -63101,7 +63101,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -63114,8 +63114,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -63125,11 +63125,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63140,11 +63140,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63155,15 +63155,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -63360,9 +63360,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -63372,7 +63372,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -63385,8 +63385,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -63396,11 +63396,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63411,11 +63411,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63426,15 +63426,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -63632,9 +63632,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -63644,7 +63644,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -63657,8 +63657,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -63668,11 +63668,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63683,11 +63683,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63698,15 +63698,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -63903,9 +63903,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -63915,7 +63915,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -63928,8 +63928,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -63939,11 +63939,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63954,11 +63954,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -63969,15 +63969,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -64173,9 +64173,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -64185,7 +64185,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -64198,8 +64198,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -64209,11 +64209,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -64224,11 +64224,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -64239,15 +64239,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -64443,9 +64443,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -64455,7 +64455,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -64468,8 +64468,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -64479,11 +64479,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -64494,11 +64494,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -64509,15 +64509,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -64714,9 +64714,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -64726,7 +64726,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -64739,8 +64739,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -64750,11 +64750,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -64765,11 +64765,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -64780,15 +64780,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -64986,9 +64986,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -64998,7 +64998,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -65011,8 +65011,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -65022,11 +65022,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65037,11 +65037,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65052,15 +65052,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -65257,9 +65257,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -65269,7 +65269,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -65282,8 +65282,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -65293,11 +65293,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65308,11 +65308,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65323,15 +65323,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -65527,9 +65527,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -65539,7 +65539,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -65552,8 +65552,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -65563,11 +65563,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65578,11 +65578,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65593,15 +65593,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -65797,9 +65797,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -65809,7 +65809,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -65822,8 +65822,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -65833,11 +65833,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65848,11 +65848,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -65863,15 +65863,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -66068,9 +66068,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -66080,7 +66080,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -66093,8 +66093,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -66104,11 +66104,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66119,11 +66119,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66134,15 +66134,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -66340,9 +66340,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -66352,7 +66352,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -66365,8 +66365,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -66376,11 +66376,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 imported LLVM 22 saturating add and fma declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66391,11 +66391,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 pure inline-PTX saturating add and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66406,15 +66406,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+              "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -66614,9 +66614,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -66626,7 +66626,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -66639,8 +66639,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -66650,11 +66650,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX fma.rn.sat.f16x2 with =r,r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7ccf304fffd63a678dec5f5775eb1fbdc60e0a295b7b2a9e878339cf8a0922a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66665,15 +66665,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ad17848cb5677cf0a68e6e3eb435b41a165f6ef1cbaf69b0fa3e2b4d8bbfdcfe. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -66858,9 +66858,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-globaltimer"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -66870,7 +66870,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -66883,8 +66883,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -66893,11 +66893,11 @@
               "targets": [
                 "sm_30"
               ],
-              "representation": "typed globaltimer declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 571-byte canonical IR with SHA-256 05f5c3e1cb6db2e75012139ce279738ff495d4073aacfd429b50bdfde6153f44.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 397d8179a2161540732e23d2358db3f5b9d44664430b730aac91c4dc83e5cded.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66908,11 +66908,11 @@
                 "sm_30",
                 "ptx32"
               ],
-              "representation": "typed globaltimer call lowering to an exact 64-bit read",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the read in 548-byte PTX with SHA-256 a78d9cae78923fa3a3be7909f47e29aedb6bfa663681173ec7a9aed44e8f4bfe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7610b182b5f89de1c960aeb1cac1a4aff14a652935f535dc6e90da6fb8a211e8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -66923,15 +66923,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining the typed globaltimer read",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_30 PTX and produced a 2,856-byte cubin with SHA-256 70b45763f48528285ad1525d565230adce99b4abc193a0f6302ca82be86024b2.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5001111b2c815febac9426aed7a4b9a51ab752be9b2d97985f60a745e49cd0a7. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -67091,9 +67091,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -67103,7 +67103,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -67116,8 +67116,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -67126,11 +67126,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed grid-dependency launch declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 73f4f353452e427acdb94ba15dbab4b8a6147137bf7e33decdbf8e41d03152e2, bitcode SHA-256 e666c4c454382bd351d83c3c8ca347bdd7d3bb9e24bb8e6643f306a4893f6ec1, and canonical IR SHA-256 7d1e8a9808ea508ca317b88b06284570bad89943a66e2dec4f7e8c7608688f94.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b0abd382879e45371de563a83219bb7d4d07b5f9f42e1be22a6f4d677203678.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -67141,11 +67141,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed griddepcontrol.launch_dependents call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted the exact instruction in PTX SHA-256 97504b6bc748ddc18d1e82bded39ecfefb288bf088e5afeca6c519f3a4f900cb.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e1f0e1a7641d1458f0fa9d84525e830a2dbb84f58fab281009a922c356df6eec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -67156,15 +67156,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced griddepcontrol.launch_dependents PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 306ce9690dd2d976940425ec7052d1f0cdb3b0cee03e8507345d328dde684796.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -67309,9 +67309,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -67321,7 +67321,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -67334,8 +67334,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -67344,11 +67344,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "typed grid-dependency wait declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 d7f02e7d83c4693e1821d37b444a9dabb20710f703938b6bf6043b5f5bf48e90, bitcode SHA-256 f1cf4c3dca46407bea36ac7957f7da7909fa6405c8f6da0fc914ab6f5277ab9a, and canonical IR SHA-256 d2d4106872992bf9ca9d7b8b147a7a7ebc6a515b5ee748b9c1cb022eacd92450.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e63a89f7a0e5ef9325815f030775cc2f52a995704e695e7f6f6bc0a9d29e7e2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -67359,11 +67359,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "typed griddepcontrol.wait call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted the exact instruction in PTX SHA-256 9b54ea0566c66befa6ff9e703c91eba251fa5c0fb9eb477a07c391fd9556045c.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 351547f724188c103868fd7cf963ba564b216c62db91f2a2a24a8685feb892a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -67374,15 +67374,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced griddepcontrol.wait PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 10bee9dcaa765674caa7cfcc9059aa263e5cf7b95fcb8fbdca3bee1f03e2b589.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -67528,9 +67528,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -67631,9 +67631,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -67734,9 +67734,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -67817,9 +67817,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-gridid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -67829,7 +67829,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -67842,8 +67842,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -67853,11 +67853,11 @@
                 "sm_30",
                 "ptx32"
               ],
-              "representation": "pure 64-bit inline PTX gridid read",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the full-width gridid read in 582-byte PTX with SHA-256 ee2f3fba8b51cd03bace8fccd9e3a76817240b7eb6e69e29943df3838b730159.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fb0578258494bd20b8dc20c6fe2243954b382980ac39ab5577a752f9af9a20e5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -67868,15 +67868,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining the 64-bit gridid read",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 39c56630ff2afdadeb521c2a56023f289a2a39f0d39b0d01e2a5b3d9171dedd9.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0236c2ff2b8a9c74dc5ef689d53ae080fbc93f5d23cee0b18105663b14080f9c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -68042,9 +68042,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-laneid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -68146,9 +68146,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-eq"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -68250,9 +68250,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-ge"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -68354,9 +68354,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-gt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -68458,9 +68458,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-le"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -68562,9 +68562,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-lt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -68717,9 +68717,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -68729,7 +68729,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -68778,8 +68778,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -68789,11 +68789,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68804,11 +68804,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68819,11 +68819,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68834,11 +68834,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68849,11 +68849,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68864,11 +68864,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68879,11 +68879,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68894,11 +68894,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68909,11 +68909,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68924,11 +68924,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68939,11 +68939,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -68954,150 +68954,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -69782,9 +69782,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -69794,7 +69794,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -69843,8 +69843,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -69854,11 +69854,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69869,11 +69869,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69884,11 +69884,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69899,11 +69899,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69914,11 +69914,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69929,11 +69929,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69944,11 +69944,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69959,11 +69959,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69974,11 +69974,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -69989,11 +69989,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70004,11 +70004,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70019,150 +70019,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -70848,9 +70848,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -70860,7 +70860,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -70909,8 +70909,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -70920,11 +70920,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70935,11 +70935,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70950,11 +70950,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70965,11 +70965,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70980,11 +70980,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -70995,11 +70995,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -71010,11 +71010,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -71025,11 +71025,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -71040,11 +71040,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -71055,11 +71055,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -71070,11 +71070,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -71085,150 +71085,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -71918,9 +71918,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -71930,7 +71930,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -71979,8 +71979,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -71990,11 +71990,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72005,11 +72005,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72020,11 +72020,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72035,11 +72035,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72050,11 +72050,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72065,11 +72065,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72080,11 +72080,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72095,11 +72095,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72110,11 +72110,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72125,11 +72125,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72140,11 +72140,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -72155,150 +72155,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -72987,9 +72987,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -72999,7 +72999,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -73048,8 +73048,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -73059,11 +73059,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73074,11 +73074,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73089,11 +73089,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73104,11 +73104,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73119,11 +73119,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73134,11 +73134,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73149,11 +73149,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73164,11 +73164,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73179,11 +73179,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73194,11 +73194,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73209,11 +73209,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -73224,150 +73224,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -74057,9 +74057,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -74069,7 +74069,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -74118,8 +74118,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -74129,11 +74129,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74144,11 +74144,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74159,11 +74159,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74174,11 +74174,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74189,11 +74189,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74204,11 +74204,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74219,11 +74219,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74234,11 +74234,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74249,11 +74249,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74264,11 +74264,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74279,11 +74279,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -74294,150 +74294,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -75121,9 +75121,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -75133,7 +75133,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -75182,8 +75182,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -75193,11 +75193,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75208,11 +75208,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75223,11 +75223,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75238,11 +75238,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75253,11 +75253,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75268,11 +75268,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75283,11 +75283,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75298,11 +75298,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75313,11 +75313,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75328,11 +75328,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75343,11 +75343,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -75358,150 +75358,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -76184,9 +76184,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -76196,7 +76196,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -76245,8 +76245,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -76256,11 +76256,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76271,11 +76271,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76286,11 +76286,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76301,11 +76301,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76316,11 +76316,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76331,11 +76331,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76346,11 +76346,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76361,11 +76361,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76376,11 +76376,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76391,11 +76391,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76406,11 +76406,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -76421,150 +76421,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -77251,9 +77251,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -77263,7 +77263,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -77312,8 +77312,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -77323,11 +77323,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77338,11 +77338,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77353,11 +77353,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77368,11 +77368,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77383,11 +77383,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77398,11 +77398,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77413,11 +77413,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77428,11 +77428,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77443,11 +77443,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77458,11 +77458,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77473,11 +77473,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -77488,150 +77488,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -78316,9 +78316,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -78328,7 +78328,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -78377,8 +78377,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -78388,11 +78388,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78403,11 +78403,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78418,11 +78418,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78433,11 +78433,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78448,11 +78448,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78463,11 +78463,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78478,11 +78478,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78493,11 +78493,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78508,11 +78508,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78523,11 +78523,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78538,11 +78538,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -78553,150 +78553,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -79387,9 +79387,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -79399,7 +79399,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -79448,8 +79448,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -79459,11 +79459,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79474,11 +79474,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79489,11 +79489,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79504,11 +79504,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79519,11 +79519,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79534,11 +79534,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79549,11 +79549,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79564,11 +79564,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79579,11 +79579,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79594,11 +79594,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79609,11 +79609,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -79624,150 +79624,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -80456,9 +80456,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -80468,7 +80468,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -80517,8 +80517,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -80528,11 +80528,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80543,11 +80543,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80558,11 +80558,11 @@
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80573,11 +80573,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80588,11 +80588,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80603,11 +80603,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80618,11 +80618,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80633,11 +80633,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80648,11 +80648,11 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80663,11 +80663,11 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80678,11 +80678,11 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -80693,150 +80693,150 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -81487,9 +81487,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -81499,7 +81499,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -81512,8 +81512,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -81522,11 +81522,11 @@
               "targets": [
                 "sm_75"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3))",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d23e19589de35127bff359facfbc56c4c16a19fd1aeef8a4568c6080ad4c1a68.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -81537,11 +81537,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3))",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x1.shared.b16 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8eb444c30aad31f7ce8a70bcd06b91bae96ddd810e5c0e0bad67d669bba76463.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -81552,15 +81552,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a38549cf45825e7cf005005dcaa96cbdff09dd83f2b672a3d5bc23b8102018e.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -81754,9 +81754,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -81766,7 +81766,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -81779,8 +81779,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -81789,11 +81789,11 @@
               "targets": [
                 "sm_75"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.trans.b16.p3(ptr addrspace(3))",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 078c3ec1a8767b2764c1a5dd2617806f672124e023a5c9fe5f3c40c6f5a36219.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -81804,11 +81804,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.trans.b16.p3(ptr addrspace(3))",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x1.trans.shared.b16 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f448dfcfc8dc1e19e552c3f601db928004ab0f095c275cf846bc4e8adc3eceb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -81819,15 +81819,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a759f49ba62f6028179716e660082e3b36877cefbe37f9ad1aa91c27aebc22c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -82024,9 +82024,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -82036,7 +82036,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -82049,8 +82049,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -82059,11 +82059,11 @@
               "targets": [
                 "sm_75"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3))",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2228a4bf5e0bbaf7685dcbd5f0e890507df94d30aa9165cbe879037b84c74a4e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82074,11 +82074,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3))",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x2.shared.b16 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c86c69e86d0084dd76afccfa7dd76416f5a906eda8f245fd5eaadfb94eeedfea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82089,15 +82089,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b42051503a9dcfd9cab9b052388ae403c55c7b81a776697ba93d8146a265b42a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -82293,9 +82293,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -82305,7 +82305,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -82318,8 +82318,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -82328,11 +82328,11 @@
               "targets": [
                 "sm_75"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3))",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b6c14f6dd0fdb52ad2e22873b19609ff47a1c8994bc5e3ce39c88bdc59a58f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82343,11 +82343,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3))",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51c2def02e859676a80a71edbbd6563d7dac75a59fb25899f3718a20705b267d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82358,15 +82358,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67149c3d0b2f90b47e4e1f3f361f390adea9ffd4186e04bafaabd93b8091168f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -82567,9 +82567,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -82579,7 +82579,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -82592,8 +82592,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -82602,11 +82602,11 @@
               "targets": [
                 "sm_75"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3))",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The recognized bare declaration materializes convergent, nocallback, memory(argmem: read), readonly, and captures(none) attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 585528599fab5e664c529894e42a2b9604bc6e503f9eefc9d2117235a0fd35fd.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82617,11 +82617,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3))",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Emitted ldmatrix.sync.aligned.m8n8.x4.shared.b16 with a four-register result struct.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3d0cc72dd11e3399dbfe72a912f451f36835e6c60ff54db70e1b6589b905dbbc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82632,15 +82632,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted x4 PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c31b7c8cbef8fc5d164d732bfa611edc166ee875735791b5446e77f385f2f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -82893,9 +82893,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -82905,7 +82905,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -82918,8 +82918,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -82928,11 +82928,11 @@
               "targets": [
                 "sm_75"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3))",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28e76c919b457758f8eeaf5f976e581db31431ae31ed1efbc30a050a31da1344.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82943,11 +82943,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3))",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c74dd5efb5771ae005406a00b33baa93333dd3fbcaa2787d0dee90957c48871f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -82958,15 +82958,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5dbfbc0015f15b42f13d4bb3e15a049ffe769e096166c94bbb0aecfaa567d43.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -83142,9 +83142,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-lg2"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -83154,7 +83154,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -83167,8 +83167,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -83178,11 +83178,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -83193,11 +83193,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -83208,15 +83208,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -83391,9 +83391,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-lg2"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -83403,7 +83403,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -83416,8 +83416,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -83427,11 +83427,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -83442,11 +83442,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -83457,15 +83457,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -83664,9 +83664,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-mapa"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -83676,7 +83676,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-cluster-memory",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-cluster-memory",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -83689,8 +83689,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -83700,11 +83700,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "one convergent LLVM kernel containing exact mapa and remote u32-load inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 690-byte source had SHA-256 ad8a94dfa560d523bfe8cf963f8779132c960e1e6e5717e080e72819936f446a; the 1,007-byte PTX had SHA-256 005a7f90c8a3bd97a99792ce8a181d43c28fb318cd20dce6068fcb349800a951.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418a46a6a0279c0e2c7391fcccae3b3bec1a660a3ffad28b1df2a7f63df297d6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -83715,15 +83715,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX retaining both cluster-memory operations",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,712-byte cubin with SHA-256 1780aee43cd691bd7e3a79f4bc4b2d163e5cdfc11e08441ae318dfb091a5057a.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bf4f5d63cf41eae708cc3087c76cbea8b9812ba26f1c915646ef4b0205c51760.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -83924,9 +83924,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -84077,9 +84077,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -84229,9 +84229,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -84381,9 +84381,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -84510,9 +84510,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -84522,7 +84522,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -84535,8 +84535,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -84546,11 +84546,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -84561,11 +84561,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -84576,26 +84576,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -84606,15 +84606,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -84808,9 +84808,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -84820,7 +84820,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -84833,8 +84833,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -84844,11 +84844,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX max.bf16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact max.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1bcd79e15bcc5f861613af0657054804f2cc98de0bc44082c84eeb6e06750d11.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -84859,15 +84859,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 37e990ab010dfd7a6aa363c1bc12abfb5d4e121bceb7b89ac8056454e95a2385.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -85043,9 +85043,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -85055,7 +85055,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -85068,8 +85068,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -85079,11 +85079,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85094,11 +85094,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85109,26 +85109,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85139,15 +85139,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -85341,9 +85341,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -85353,7 +85353,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -85366,8 +85366,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -85377,11 +85377,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX max.f16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4070d164dfb7ec86737afcc51fdef8e9452193df799721221b988492eab3188.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85392,15 +85392,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined three-form sm_80 packed-f16x2 upper probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9dc936add8025c51ff86d52cb72929dbbd1197519c0ffed33b53ba5eae61c7c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -85576,9 +85576,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -85588,7 +85588,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -85601,8 +85601,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -85612,11 +85612,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85627,11 +85627,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85642,26 +85642,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85672,15 +85672,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -85875,9 +85875,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -85887,7 +85887,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -85900,8 +85900,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -85911,11 +85911,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85926,11 +85926,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85941,26 +85941,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -85971,15 +85971,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -86174,9 +86174,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -86186,7 +86186,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -86199,8 +86199,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -86210,11 +86210,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86225,11 +86225,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86240,26 +86240,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86270,15 +86270,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -86474,9 +86474,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -86486,7 +86486,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -86499,8 +86499,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -86510,11 +86510,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86525,11 +86525,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86540,26 +86540,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86570,15 +86570,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -86774,9 +86774,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -86786,7 +86786,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -86799,8 +86799,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -86810,11 +86810,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86825,11 +86825,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86840,26 +86840,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -86870,15 +86870,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -87076,9 +87076,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -87088,7 +87088,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -87101,8 +87101,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -87112,11 +87112,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87127,11 +87127,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87142,26 +87142,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87172,15 +87172,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -87378,9 +87378,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -87390,7 +87390,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -87403,8 +87403,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -87414,11 +87414,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87429,11 +87429,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87444,26 +87444,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87474,15 +87474,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -87680,9 +87680,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -87692,7 +87692,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -87705,8 +87705,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -87716,11 +87716,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87731,11 +87731,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87746,26 +87746,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -87776,15 +87776,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -87981,9 +87981,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -87993,7 +87993,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -88006,8 +88006,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -88017,11 +88017,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88032,11 +88032,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88047,26 +88047,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88077,15 +88077,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -88282,9 +88282,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -88294,7 +88294,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -88307,8 +88307,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -88318,11 +88318,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88333,11 +88333,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88348,26 +88348,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88378,15 +88378,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -88583,9 +88583,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -88595,7 +88595,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -88608,8 +88608,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -88619,11 +88619,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88634,11 +88634,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88649,26 +88649,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88679,15 +88679,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -88882,9 +88882,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -88894,7 +88894,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -88907,8 +88907,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -88918,11 +88918,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88933,11 +88933,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88948,26 +88948,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -88978,15 +88978,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -89181,9 +89181,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -89193,7 +89193,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -89206,8 +89206,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -89217,11 +89217,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89232,11 +89232,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89247,26 +89247,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89277,15 +89277,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -89480,9 +89480,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -89492,7 +89492,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -89505,8 +89505,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -89516,11 +89516,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89531,11 +89531,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89546,26 +89546,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89576,15 +89576,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -89779,9 +89779,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -89791,7 +89791,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -89804,8 +89804,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -89815,11 +89815,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89830,11 +89830,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89845,26 +89845,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -89875,15 +89875,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -90080,9 +90080,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -90092,7 +90092,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -90105,8 +90105,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -90116,11 +90116,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90131,11 +90131,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90146,26 +90146,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90176,15 +90176,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -90381,9 +90381,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -90393,7 +90393,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -90406,8 +90406,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -90417,11 +90417,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90432,11 +90432,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90447,26 +90447,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90477,15 +90477,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -90682,9 +90682,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -90694,7 +90694,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -90707,8 +90707,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -90718,11 +90718,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90733,11 +90733,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90748,26 +90748,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -90778,15 +90778,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -90983,9 +90983,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -90995,7 +90995,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -91008,8 +91008,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -91019,11 +91019,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91034,11 +91034,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91049,26 +91049,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91079,15 +91079,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -91255,9 +91255,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -91267,7 +91267,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -91280,8 +91280,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -91291,11 +91291,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX max.relu.s16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 692-byte PTX module has SHA-256 8d5d0d27786cd1453843d71929ee823f7f864cf2bb31388b3f35924e5a78239b. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5fc39021e5a2dd08e1b4c600bd65bfbd719061fd7182b4f87325793bcfeca3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91393,9 +91393,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -91405,7 +91405,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -91418,8 +91418,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -91429,11 +91429,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX max.relu.s32 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 676-byte PTX module has SHA-256 34e11547bce5d21764f671e36d47c30191aa492a57f253068f1a6f9ccb47d4cc. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 47ca916b21f4f81062a57a273ed3a2574d88dc7961f3ed061ee81135c02019a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91531,9 +91531,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -91543,7 +91543,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -91556,8 +91556,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -91567,11 +91567,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX max.s16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 3a29f56af95dac7f78fa65c80cda474d1b2f71846240c18e28d1b227bed13bbe. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 019612a6c0a895437ac9e29439ca4404a44adf617072fb390270e41680557112.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91668,9 +91668,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -91680,7 +91680,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -91693,8 +91693,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -91704,11 +91704,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX max.u16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 af376fac111934ef1cc375e57377b0eef4fd31e226db9eed52811fe89f924953. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 493b5e74e07a0232486d2df12ed0c80ed712745b36b51c17cfec678c28fdcb97.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91834,9 +91834,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -91846,7 +91846,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -91859,8 +91859,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -91870,11 +91870,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91885,11 +91885,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91900,26 +91900,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -91930,15 +91930,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -92134,9 +92134,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -92146,7 +92146,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -92159,8 +92159,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -92170,11 +92170,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92185,11 +92185,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92200,26 +92200,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92230,15 +92230,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -92434,9 +92434,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -92446,7 +92446,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -92459,8 +92459,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -92470,11 +92470,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92485,11 +92485,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92500,26 +92500,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92530,15 +92530,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -92734,9 +92734,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -92746,7 +92746,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -92759,8 +92759,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -92770,11 +92770,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92785,11 +92785,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92800,26 +92800,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -92830,15 +92830,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -93034,9 +93034,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -93046,7 +93046,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -93059,8 +93059,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -93070,11 +93070,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93085,11 +93085,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93100,26 +93100,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93130,15 +93130,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -93329,9 +93329,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -93341,7 +93341,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -93354,8 +93354,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -93365,11 +93365,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.mbarrier.arrive.shared declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7df70d42d790f340f00c6f29da571b9b809bee9f1d093b742ce1a0e421272f46.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93380,11 +93380,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed shared-memory mbarrier arrival",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact mbarrier.arrive.shared.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d140364a2d31493758ea19f5e725bbfea20f14d6ce43489b079bb4789e428ce.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93395,15 +93395,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57bdb80f49dc0b09308813d01303ede4a2ba34e99e63aaecd6bea6813137fd7e.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -93545,9 +93545,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -93557,7 +93557,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -93570,8 +93570,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -93581,11 +93581,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93596,15 +93596,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX 8.0 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -93772,9 +93772,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -93784,7 +93784,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -93797,8 +93797,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -93808,11 +93808,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -93823,15 +93823,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX 8.0 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -94003,9 +94003,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -94015,7 +94015,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -94028,8 +94028,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -94039,11 +94039,11 @@
                 "sm_90",
                 "ptx86"
               ],
-              "representation": "generated cluster-scope inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94054,15 +94054,15 @@
                 "sm_90",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX 8.6 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -94241,9 +94241,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -94253,7 +94253,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -94266,8 +94266,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -94277,11 +94277,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.mbarrier.arrive.noComplete.shared declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the counted shared-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3eb1f951460f6bc67c6055899542860e1010e5db5406f5e2f9accee3cbc170c2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94292,11 +94292,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed counted shared-memory mbarrier no-complete arrival",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact mbarrier.arrive.noComplete.shared.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 174b9c4a382274b14d72edd3fd22bdc8d2486f5cb87d6c8bb4a48e7bc66b7000.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94307,15 +94307,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b579f9834ca9bfb229788f02fc0ce145ba59fac61cbb61036bc8c552cb4bbb7a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -94488,9 +94488,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-init"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -94500,7 +94500,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -94513,8 +94513,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -94524,11 +94524,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.mbarrier.init.shared declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28fcac92a1ebd5fca5408ed34c7aa64ba3546bbe5714355a7ebfd14b92f760e3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94539,11 +94539,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed shared-memory mbarrier initialization",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact mbarrier.init.shared.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 50ba004fcfb85da96f6f3347966e4704dda4443ace6b57676924bd528c08bf0b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94554,15 +94554,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ab45efc26f5f14c513e6ab0ed925ffddbdfc1cb0e890ef4f445a40e88acebc9.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -94732,9 +94732,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-inval"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -94744,7 +94744,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -94757,8 +94757,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -94768,11 +94768,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed llvm.nvvm.mbarrier.inval.shared declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the declaration with convergent, nocallback, and argument-memory write attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76483bcaa272bf6c2c1688ff61ce630eba2cbc651af55d047b7048e43352adb7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94783,11 +94783,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed shared-memory mbarrier invalidation",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact mbarrier.inval.shared.b64 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4effb270bc9d6ccd5d7c24a4518d532eb913541a234db702e0d50b7c4f7f0774.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -94798,15 +94798,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 453f943cafeb60e5761ed5f0867ebb1e2519a3dd314eb9b0dc4b5e6ff6fbe65c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -94976,9 +94976,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -94988,7 +94988,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -95001,8 +95001,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -95012,11 +95012,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent side-effecting inline PTX mbarrier test-wait",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend retained exact mbarrier.test_wait.shared.b64 PTX and the predicate-to-i1 adapter.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6a81cac5ab0bf8f94c54775467e1df856112351608671f11a4e8e63ffb75f00.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -95027,15 +95027,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df51ed0e79da4ce10d9298aa7e4fc3db4b75355044d51f95f92694acde7f9e50.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -95212,9 +95212,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -95224,7 +95224,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -95237,8 +95237,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -95248,11 +95248,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact instruction and predicate adapter at the route floor.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5dbfe0da451289673522a20ec37fd0664d5276bf7760b0d8a8484d613e82ca01.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -95263,15 +95263,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX 7.8 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24bb1e758b565ffc3df774b43f1cdc6f7b26a5f827d1add384e6ffcd78747080.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -95439,9 +95439,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -95451,7 +95451,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -95464,8 +95464,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -95475,11 +95475,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact instruction and predicate adapter at the route floor.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5dbfe0da451289673522a20ec37fd0664d5276bf7760b0d8a8484d613e82ca01.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -95490,15 +95490,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX 7.8 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24bb1e758b565ffc3df774b43f1cdc6f7b26a5f827d1add384e6ffcd78747080.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -95667,9 +95667,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -95679,7 +95679,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -95692,8 +95692,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -95703,11 +95703,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "generated extended-mbarrier inline-PTX probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -95718,15 +95718,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "LLVM-produced PTX 8.0 modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -95907,9 +95907,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -95919,7 +95919,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -95932,8 +95932,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -95943,11 +95943,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -95958,11 +95958,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -95973,26 +95973,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -96003,15 +96003,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -96205,9 +96205,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -96217,7 +96217,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -96230,8 +96230,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -96241,11 +96241,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX min.bf16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact min.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f59f875a1a686231cf0e0c09cd8ea1d7107aaf7ba547e54f0847ba4d71726b27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -96256,15 +96256,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51930c796a141b52e8382f91a54de007b9908945d7bdddf8085b3c9e4575557a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -96440,9 +96440,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -96452,7 +96452,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -96465,8 +96465,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -96476,11 +96476,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -96491,11 +96491,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -96506,26 +96506,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -96536,15 +96536,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -96738,9 +96738,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -96750,7 +96750,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -96763,8 +96763,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -96774,11 +96774,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX min.f16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d54f22a9e49d8c8e53739fb3942a54749c50fe7749970cd67c4c77932fc1e11.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -96789,15 +96789,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined three-form sm_80 packed-f16x2 upper probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34e9c8ebf06b8ee9b1df5eda184d80fb12116914e533e5f11ecf26f337751abc.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -96973,9 +96973,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -96985,7 +96985,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -96998,8 +96998,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -97009,11 +97009,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97024,11 +97024,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97039,26 +97039,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97069,15 +97069,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -97272,9 +97272,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -97284,7 +97284,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -97297,8 +97297,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -97308,11 +97308,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97323,11 +97323,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97338,26 +97338,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97368,15 +97368,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -97571,9 +97571,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -97583,7 +97583,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -97596,8 +97596,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -97607,11 +97607,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97622,11 +97622,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97637,26 +97637,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97667,15 +97667,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -97871,9 +97871,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -97883,7 +97883,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -97896,8 +97896,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -97907,11 +97907,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97922,11 +97922,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97937,26 +97937,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -97967,15 +97967,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -98171,9 +98171,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -98183,7 +98183,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -98196,8 +98196,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -98207,11 +98207,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98222,11 +98222,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98237,26 +98237,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98267,15 +98267,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -98473,9 +98473,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -98485,7 +98485,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -98498,8 +98498,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -98509,11 +98509,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98524,11 +98524,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98539,26 +98539,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98569,15 +98569,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -98775,9 +98775,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -98787,7 +98787,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -98800,8 +98800,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -98811,11 +98811,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98826,11 +98826,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98841,26 +98841,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -98871,15 +98871,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -99077,9 +99077,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -99089,7 +99089,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -99102,8 +99102,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -99113,11 +99113,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99128,11 +99128,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99143,26 +99143,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99173,15 +99173,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -99378,9 +99378,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -99390,7 +99390,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -99403,8 +99403,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -99414,11 +99414,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99429,11 +99429,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99444,26 +99444,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99474,15 +99474,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -99679,9 +99679,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -99691,7 +99691,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -99704,8 +99704,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -99715,11 +99715,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99730,11 +99730,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99745,26 +99745,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -99775,15 +99775,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -99980,9 +99980,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -99992,7 +99992,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -100005,8 +100005,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -100016,11 +100016,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100031,11 +100031,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100046,26 +100046,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100076,15 +100076,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -100279,9 +100279,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -100291,7 +100291,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -100304,8 +100304,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -100315,11 +100315,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100330,11 +100330,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100345,26 +100345,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100375,15 +100375,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -100578,9 +100578,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -100590,7 +100590,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -100603,8 +100603,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -100614,11 +100614,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100629,11 +100629,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100644,26 +100644,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100674,15 +100674,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -100877,9 +100877,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -100889,7 +100889,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -100902,8 +100902,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -100913,11 +100913,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100928,11 +100928,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 typed llvm.nvvm min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100943,26 +100943,26 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "typed LLVM PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "8 pure production inline-PTX min/max calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -100973,15 +100973,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "production inline PTX 7.0 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -101176,9 +101176,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -101188,7 +101188,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -101201,8 +101201,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -101212,11 +101212,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101227,11 +101227,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101242,26 +101242,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101272,15 +101272,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -101477,9 +101477,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -101489,7 +101489,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -101502,8 +101502,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -101513,11 +101513,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101528,11 +101528,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101543,26 +101543,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101573,15 +101573,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -101778,9 +101778,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -101790,7 +101790,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -101803,8 +101803,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -101814,11 +101814,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101829,11 +101829,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101844,26 +101844,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -101874,15 +101874,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -102079,9 +102079,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -102091,7 +102091,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -102104,8 +102104,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -102115,11 +102115,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102130,11 +102130,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102145,26 +102145,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102175,15 +102175,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -102380,9 +102380,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -102392,7 +102392,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -102405,8 +102405,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -102416,11 +102416,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102431,11 +102431,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102446,26 +102446,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102476,15 +102476,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -102652,9 +102652,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -102664,7 +102664,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -102677,8 +102677,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -102688,11 +102688,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX min.relu.s16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 692-byte PTX module has SHA-256 c10411ef8355e759ea266343683a7f30e971822dcdf3f47eb91c74bc44a92c95. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ccd77090590ed5077784a28895d4db44c0c3601913263b77b3afedf890a94db1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102790,9 +102790,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -102802,7 +102802,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -102815,8 +102815,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -102826,11 +102826,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX min.relu.s32 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 676-byte PTX module has SHA-256 e835563e9676e88090632ce5c529840da1603ff6e65b3b88f2d722c626bd30d8. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05737e13d7ecd4d7d29377123087680401a96d0f6423842f9eb0c570c819a156.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -102928,9 +102928,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -102940,7 +102940,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -102953,8 +102953,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -102964,11 +102964,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX min.s16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 2916685ee0644fd8029821bab02c8cb4155b0ed1f3cce20b10c9dbe58edacf0a. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2b9b8cd02f5dbb4b21bbe79e50ef96e87c4ff8a61d113774a049c1df97e2d645.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103065,9 +103065,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -103077,7 +103077,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -103090,8 +103090,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -103101,11 +103101,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "pure inline PTX min.u16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 1d08c966ac1bea96a509823b5dc389325f2ac8a059d04619a8ff3e428e4f5d45. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\".",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bc038f0362d6863309339980bbb161649c4774ba6fc3e43058645c320efd1724.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103231,9 +103231,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -103243,7 +103243,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -103256,8 +103256,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -103267,11 +103267,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103282,11 +103282,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103297,26 +103297,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103327,15 +103327,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -103531,9 +103531,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -103543,7 +103543,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -103556,8 +103556,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -103567,11 +103567,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103582,11 +103582,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103597,26 +103597,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103627,15 +103627,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -103831,9 +103831,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -103843,7 +103843,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -103856,8 +103856,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -103867,11 +103867,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103882,11 +103882,11 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103897,26 +103897,26 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "typed LLVM PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -103927,15 +103927,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX 7.2 at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+              "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -104131,9 +104131,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -104143,7 +104143,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -104156,8 +104156,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -104167,11 +104167,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104182,11 +104182,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104197,26 +104197,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104227,15 +104227,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -104431,9 +104431,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_86",
@@ -104443,7 +104443,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.2",
             "hardware": {
@@ -104456,8 +104456,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -104467,11 +104467,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 imported typed LLVM declarations and calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104482,11 +104482,11 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104497,26 +104497,26 @@
                 "sm_86",
                 "ptx78"
               ],
-              "representation": "typed LLVM PTX 7.8 carrier",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104527,15 +104527,15 @@
                 "sm_86",
                 "ptx72"
               ],
-              "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+              "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -104747,9 +104747,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -104759,7 +104759,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -104772,8 +104772,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -104783,11 +104783,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k128 B1 AND-popcount MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0786eb2ac48a149064adcb908b142249e47b70b7dc6b77221c691534d54f8363.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -104798,15 +104798,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for m16n8k128 B1 AND-popcount MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d7868cbe80c20cad2dafd27dc6429792a50fd3b4efb0da2dd55b4d836fedaef9.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -105022,9 +105022,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -105034,7 +105034,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -105047,8 +105047,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -105058,11 +105058,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k128 B1 XOR-popcount MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4fc3a76fa4f4f3118a8078863963902f0b33ebeecb269fccefaf2d785a9171f4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -105073,15 +105073,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "exact-floor PTX for m16n8k128 B1 XOR-popcount MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_80 floors.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 66efb066dd43977add4c7b6812754fc0affb5166a6dee22b060759d4eac2e5ab.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -105295,9 +105295,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -105307,7 +105307,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -105320,8 +105320,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -105334,8 +105334,8 @@
               "representation": "m16n8k16 F16 lower-hardware boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the form below sm_80; the diagnostic SHA-256 is f6582289bf7df33af5bf1c016ce0f73335a4a37ca257c42d201fe9cfefbe134c.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the form below sm_80; the diagnostic SHA-256 is f6582289bf7df33af5bf1c016ce0f73335a4a37ca257c42d201fe9cfefbe134c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -105346,11 +105346,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five typed LLVM declarations and duplicate-call probes",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -105361,11 +105361,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five convergent side-effecting register-only inline PTX calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -105376,15 +105376,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "aggregate inline-PTX MMA fixture at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -105619,9 +105619,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -105631,7 +105631,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -105644,8 +105644,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -105655,11 +105655,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for BF16 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact BF16 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0a3bd41a37b501ef0419e036d4a2325f28dee3d5d47c050ab1af6d52b9904385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -105670,15 +105670,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b0d9dbc95cc97fb422f1b7b9e39d1d68c866b4ffcf48c9a32f8fb917454bb82.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -105898,9 +105898,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -105910,7 +105910,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -105923,8 +105923,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -105934,11 +105934,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for F16 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact F16 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cb21825affcdfcb250dea8bfead8827d200376f1b6dfb69b7659d670b4fe0a8d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -105949,15 +105949,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d25700d38dead0336c44add15c4dc07020aab017e285b96e0e94bf091e743e4e.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -106163,9 +106163,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -106175,7 +106175,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -106188,8 +106188,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -106199,11 +106199,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and representative duplicate calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -106214,11 +106214,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed and convergent inline standard FP8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -106229,15 +106229,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F16 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -106247,8 +106247,8 @@
               "representation": "k16 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -106262,8 +106262,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -106519,9 +106519,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -106531,7 +106531,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -106544,8 +106544,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -106555,11 +106555,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and representative duplicate calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -106570,11 +106570,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed and convergent inline standard FP8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -106585,15 +106585,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F16 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -106603,8 +106603,8 @@
               "representation": "k16 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -106618,8 +106618,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -106875,9 +106875,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -106887,7 +106887,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -106900,8 +106900,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -106911,11 +106911,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and representative duplicate calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -106926,11 +106926,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed and convergent inline standard FP8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -106941,15 +106941,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F16 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -106959,8 +106959,8 @@
               "representation": "k16 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -106974,8 +106974,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -107231,9 +107231,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -107243,7 +107243,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -107256,8 +107256,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -107267,11 +107267,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and representative duplicate calls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -107282,11 +107282,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed and convergent inline standard FP8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -107297,15 +107297,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F16 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -107315,8 +107315,8 @@
               "representation": "k16 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -107330,8 +107330,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -107595,9 +107595,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -107607,7 +107607,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -107620,8 +107620,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -107631,11 +107631,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -107646,11 +107646,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k16 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -107661,15 +107661,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -107679,8 +107679,8 @@
               "representation": "k16 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k16 FP8 MMA below PTX 8.7.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -107694,8 +107694,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -107959,9 +107959,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -107971,7 +107971,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -107984,8 +107984,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -107995,11 +107995,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -108010,11 +108010,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k16 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -108025,15 +108025,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -108043,8 +108043,8 @@
               "representation": "k16 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k16 FP8 MMA below PTX 8.7.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -108058,8 +108058,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -108323,9 +108323,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -108335,7 +108335,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -108348,8 +108348,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -108359,11 +108359,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -108374,11 +108374,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k16 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -108389,15 +108389,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -108407,8 +108407,8 @@
               "representation": "k16 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k16 FP8 MMA below PTX 8.7.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -108422,8 +108422,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -108687,9 +108687,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -108699,7 +108699,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -108712,8 +108712,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -108723,11 +108723,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -108738,11 +108738,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k16 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -108753,15 +108753,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k16 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -108771,8 +108771,8 @@
               "representation": "k16 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k16 FP8 MMA below PTX 8.7.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -108786,8 +108786,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -109051,9 +109051,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -109063,7 +109063,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -109076,8 +109076,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -109087,11 +109087,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k16 signed INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping s8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a48e36d7da44b51913001f1eb7d4d1a4c3fe187828db3a37451079cfb79bd041.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -109102,15 +109102,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3f2318b8382d87d1f67a2cc34a6842dbb8d910a85130f0d3d991690b39d6f1c2.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -109324,9 +109324,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -109336,7 +109336,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -109349,8 +109349,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -109360,11 +109360,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k16 signed INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite s8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32f364dd111daf1d85405149c434903c9ba832ee2d5d6e3067ab468e61bcd713.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -109375,15 +109375,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f136aaf2137b3c1f962fdcdbeaf7f7ed4ad688be4fb5cd55dadaf7ac100fc153.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -109598,9 +109598,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -109610,7 +109610,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -109623,8 +109623,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -109634,11 +109634,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping s8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c7ad5689bb1dc7d9f6e66a3b7024246cecd620c8499b9468999c5efc7cf4ef38.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -109649,15 +109649,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2318600ab58b33347303547db32bfbfdccea5210d1401891255d10c18a8d89cd.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -109871,9 +109871,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -109883,7 +109883,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -109896,8 +109896,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -109907,11 +109907,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite s8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 516ff5a0c66113c514e59c0272876c73360c509581521ac8ef9c6c7b1a02c82e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -109922,15 +109922,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b15a9cd586a5fbde549415441be60eae06e57d99604ba2350ca476c0066f6a1a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -110145,9 +110145,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -110157,7 +110157,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -110170,8 +110170,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -110181,11 +110181,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k16 unsigned INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping u8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 db2d33329053739e1f39831d56e2d27e487dd334679bb09de2f5c8f80eb1de65.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -110196,15 +110196,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0a674a423572b924a7f2e6dd0a2c69d601e5d2c4682dfa1414103e7045a686c6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -110418,9 +110418,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -110430,7 +110430,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -110443,8 +110443,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -110454,11 +110454,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping u8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9921a009519c555a2adb72baf0e136405ef6b850b9a19d83f10e7d0b3d763f71.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -110469,15 +110469,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 969755d83267f71bbb72953caff716f8eae65ee1d14c4a293879a0a21338c337.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -110691,9 +110691,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -110703,7 +110703,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -110716,8 +110716,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -110727,11 +110727,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite u8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab5b0e911466d87855d48864c54ff2da53c2e4e5b940b1c0812b378ea453e585.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -110742,15 +110742,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a300eed5eea4deeaf14e77ce212087a11d0a912a110956a8f98d7ef0219018cf.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -110965,9 +110965,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -110977,7 +110977,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -110990,8 +110990,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -111001,11 +111001,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k16 unsigned INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite u8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 332115a18b93fa8d2c7916c9fe028636dde30234d0b197244146f55eca052148.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111016,15 +111016,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8301b6cd488dea73b3278090a9a30910809f2afba5c3e9b58b21a1d84cdca8bd.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -111246,9 +111246,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -111258,7 +111258,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -111271,8 +111271,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -111282,11 +111282,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k256 B1 AND-popcount MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b08bb4ac9ef5d95f902982d6d6e4da9bc6e758f02a86e4f0da412b6dd22f561c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111297,15 +111297,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for m16n8k256 B1 AND-popcount MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd92d37a566e4ee028ef24ef3bc175b98a0ade64ebf874e29c45a409119ee2bb.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -111527,9 +111527,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -111539,7 +111539,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -111552,8 +111552,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -111563,11 +111563,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k256 B1 XOR-popcount MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e760db526d0a3aef6cdfb3a2032edb8ef925f23d7c046b15eb455e304c0a66b2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111578,15 +111578,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "exact-floor PTX for m16n8k256 B1 XOR-popcount MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_80 floors.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f577f397dc588d57cf9709716a198f6b17f1d4f6fa071c808bcaff2ddb58b8dc.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -111811,9 +111811,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -111823,7 +111823,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -111848,8 +111848,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -111859,11 +111859,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111874,11 +111874,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111889,11 +111889,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111904,26 +111904,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111934,26 +111934,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111964,26 +111964,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -111994,15 +111994,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -112012,8 +112012,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112027,8 +112027,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112042,8 +112042,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112057,8 +112057,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112072,8 +112072,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112087,8 +112087,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112102,8 +112102,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112508,9 +112508,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -112520,7 +112520,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -112545,8 +112545,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -112556,11 +112556,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -112571,11 +112571,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -112586,11 +112586,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -112601,26 +112601,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -112631,26 +112631,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -112661,26 +112661,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -112691,15 +112691,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -112709,8 +112709,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112724,8 +112724,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112739,8 +112739,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112754,8 +112754,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112769,8 +112769,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112784,8 +112784,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -112799,8 +112799,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113205,9 +113205,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -113217,7 +113217,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -113242,8 +113242,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -113253,11 +113253,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113268,11 +113268,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113283,11 +113283,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113298,26 +113298,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113328,26 +113328,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113358,26 +113358,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113388,15 +113388,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -113406,8 +113406,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113421,8 +113421,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113436,8 +113436,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113451,8 +113451,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113466,8 +113466,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113481,8 +113481,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113496,8 +113496,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -113902,9 +113902,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -113914,7 +113914,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -113939,8 +113939,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -113950,11 +113950,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113965,11 +113965,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113980,11 +113980,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -113995,26 +113995,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114025,26 +114025,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114055,26 +114055,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114085,15 +114085,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -114103,8 +114103,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114118,8 +114118,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114133,8 +114133,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114148,8 +114148,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114163,8 +114163,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114178,8 +114178,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114193,8 +114193,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114599,9 +114599,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -114611,7 +114611,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -114636,8 +114636,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -114647,11 +114647,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114662,11 +114662,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114677,11 +114677,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114692,26 +114692,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114722,26 +114722,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114752,26 +114752,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -114782,15 +114782,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -114800,8 +114800,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114815,8 +114815,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114830,8 +114830,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114845,8 +114845,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114860,8 +114860,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114875,8 +114875,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -114890,8 +114890,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115296,9 +115296,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -115308,7 +115308,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -115333,8 +115333,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -115344,11 +115344,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -115359,11 +115359,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -115374,11 +115374,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -115389,26 +115389,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -115419,26 +115419,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -115449,26 +115449,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -115479,15 +115479,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -115497,8 +115497,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115512,8 +115512,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115527,8 +115527,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115542,8 +115542,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115557,8 +115557,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115572,8 +115572,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115587,8 +115587,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -115993,9 +115993,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -116005,7 +116005,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -116030,8 +116030,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -116041,11 +116041,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116056,11 +116056,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116071,11 +116071,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116086,26 +116086,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116116,26 +116116,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116146,26 +116146,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116176,15 +116176,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -116194,8 +116194,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116209,8 +116209,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116224,8 +116224,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116239,8 +116239,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116254,8 +116254,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116269,8 +116269,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116284,8 +116284,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116690,9 +116690,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -116702,7 +116702,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -116727,8 +116727,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -116738,11 +116738,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116753,11 +116753,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116768,11 +116768,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116783,26 +116783,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116813,26 +116813,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116843,26 +116843,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -116873,15 +116873,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -116891,8 +116891,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116906,8 +116906,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116921,8 +116921,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116936,8 +116936,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116951,8 +116951,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116966,8 +116966,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -116981,8 +116981,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117387,9 +117387,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -117399,7 +117399,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -117424,8 +117424,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -117435,11 +117435,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -117450,11 +117450,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -117465,11 +117465,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -117480,26 +117480,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -117510,26 +117510,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -117540,26 +117540,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -117570,15 +117570,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -117588,8 +117588,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117603,8 +117603,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117618,8 +117618,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117633,8 +117633,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117648,8 +117648,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117663,8 +117663,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -117678,8 +117678,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118084,9 +118084,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -118096,7 +118096,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -118121,8 +118121,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -118132,11 +118132,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118147,11 +118147,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118162,11 +118162,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118177,26 +118177,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118207,26 +118207,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118237,26 +118237,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118267,15 +118267,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -118285,8 +118285,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118300,8 +118300,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118315,8 +118315,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118330,8 +118330,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118345,8 +118345,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118360,8 +118360,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118375,8 +118375,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118781,9 +118781,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -118793,7 +118793,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -118818,8 +118818,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -118829,11 +118829,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118844,11 +118844,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118859,11 +118859,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118874,26 +118874,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118904,26 +118904,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118934,26 +118934,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -118964,15 +118964,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -118982,8 +118982,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -118997,8 +118997,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119012,8 +119012,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119027,8 +119027,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119042,8 +119042,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119057,8 +119057,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119072,8 +119072,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119478,9 +119478,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -119490,7 +119490,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -119515,8 +119515,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -119526,11 +119526,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -119541,11 +119541,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -119556,11 +119556,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -119571,26 +119571,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -119601,26 +119601,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -119631,26 +119631,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -119661,15 +119661,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -119679,8 +119679,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119694,8 +119694,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119709,8 +119709,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119724,8 +119724,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119739,8 +119739,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119754,8 +119754,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -119769,8 +119769,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120175,9 +120175,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -120187,7 +120187,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -120212,8 +120212,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -120223,11 +120223,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120238,11 +120238,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120253,11 +120253,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120268,26 +120268,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120298,26 +120298,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120328,26 +120328,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120358,15 +120358,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -120376,8 +120376,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120391,8 +120391,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120406,8 +120406,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120421,8 +120421,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120436,8 +120436,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120451,8 +120451,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120466,8 +120466,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -120872,9 +120872,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -120884,7 +120884,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -120909,8 +120909,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -120920,11 +120920,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120935,11 +120935,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120950,11 +120950,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120965,26 +120965,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -120995,26 +120995,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121025,26 +121025,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121055,15 +121055,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -121073,8 +121073,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121088,8 +121088,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121103,8 +121103,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121118,8 +121118,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121133,8 +121133,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121148,8 +121148,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121163,8 +121163,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121569,9 +121569,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -121581,7 +121581,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -121606,8 +121606,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -121617,11 +121617,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121632,11 +121632,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121647,11 +121647,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121662,26 +121662,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121692,26 +121692,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121722,26 +121722,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -121752,15 +121752,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -121770,8 +121770,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121785,8 +121785,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121800,8 +121800,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121815,8 +121815,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121830,8 +121830,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121845,8 +121845,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -121860,8 +121860,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122266,9 +122266,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -122278,7 +122278,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -122303,8 +122303,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -122314,11 +122314,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -122329,11 +122329,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -122344,11 +122344,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -122359,26 +122359,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -122389,26 +122389,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -122419,26 +122419,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -122449,15 +122449,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -122467,8 +122467,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122482,8 +122482,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122497,8 +122497,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122512,8 +122512,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122527,8 +122527,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122542,8 +122542,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122557,8 +122557,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -122963,9 +122963,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -122975,7 +122975,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -123000,8 +123000,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -123011,11 +123011,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123026,11 +123026,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123041,11 +123041,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123056,26 +123056,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123086,26 +123086,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123116,26 +123116,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123146,15 +123146,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -123164,8 +123164,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123179,8 +123179,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123194,8 +123194,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123209,8 +123209,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123224,8 +123224,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123239,8 +123239,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123254,8 +123254,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123660,9 +123660,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -123672,7 +123672,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -123697,8 +123697,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -123708,11 +123708,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123723,11 +123723,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123738,11 +123738,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123753,26 +123753,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123783,26 +123783,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123813,26 +123813,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -123843,15 +123843,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -123861,8 +123861,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123876,8 +123876,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123891,8 +123891,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123906,8 +123906,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123921,8 +123921,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123936,8 +123936,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -123951,8 +123951,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124357,9 +124357,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -124369,7 +124369,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -124394,8 +124394,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -124405,11 +124405,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -124420,11 +124420,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -124435,11 +124435,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -124450,26 +124450,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -124480,26 +124480,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -124510,26 +124510,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -124540,15 +124540,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -124558,8 +124558,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124573,8 +124573,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124588,8 +124588,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124603,8 +124603,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124618,8 +124618,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124633,8 +124633,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -124648,8 +124648,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125054,9 +125054,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -125066,7 +125066,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -125091,8 +125091,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -125102,11 +125102,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125117,11 +125117,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125132,11 +125132,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125147,26 +125147,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125177,26 +125177,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125207,26 +125207,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125237,15 +125237,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -125255,8 +125255,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125270,8 +125270,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125285,8 +125285,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125300,8 +125300,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125315,8 +125315,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125330,8 +125330,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125345,8 +125345,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125751,9 +125751,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -125763,7 +125763,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -125788,8 +125788,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -125799,11 +125799,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125814,11 +125814,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125829,11 +125829,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125844,26 +125844,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125874,26 +125874,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125904,26 +125904,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -125934,15 +125934,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -125952,8 +125952,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125967,8 +125967,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125982,8 +125982,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -125997,8 +125997,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126012,8 +126012,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126027,8 +126027,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126042,8 +126042,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126448,9 +126448,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -126460,7 +126460,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -126485,8 +126485,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -126496,11 +126496,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -126511,11 +126511,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -126526,11 +126526,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -126541,26 +126541,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -126571,26 +126571,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -126601,26 +126601,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -126631,15 +126631,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -126649,8 +126649,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126664,8 +126664,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126679,8 +126679,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126694,8 +126694,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126709,8 +126709,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126724,8 +126724,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -126739,8 +126739,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127145,9 +127145,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -127157,7 +127157,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -127182,8 +127182,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -127193,11 +127193,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127208,11 +127208,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127223,11 +127223,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127238,26 +127238,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127268,26 +127268,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127298,26 +127298,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127328,15 +127328,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -127346,8 +127346,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127361,8 +127361,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127376,8 +127376,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127391,8 +127391,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127406,8 +127406,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127421,8 +127421,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127436,8 +127436,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -127842,9 +127842,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -127854,7 +127854,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -127879,8 +127879,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -127890,11 +127890,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127905,11 +127905,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127920,11 +127920,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127935,26 +127935,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127965,26 +127965,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -127995,26 +127995,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128025,15 +128025,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -128043,8 +128043,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128058,8 +128058,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128073,8 +128073,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128088,8 +128088,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128103,8 +128103,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128118,8 +128118,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128133,8 +128133,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128539,9 +128539,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -128551,7 +128551,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -128576,8 +128576,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -128587,11 +128587,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128602,11 +128602,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128617,11 +128617,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128632,26 +128632,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128662,26 +128662,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128692,26 +128692,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -128722,15 +128722,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -128740,8 +128740,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128755,8 +128755,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128770,8 +128770,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128785,8 +128785,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128800,8 +128800,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128815,8 +128815,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -128830,8 +128830,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129244,9 +129244,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -129256,7 +129256,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -129281,8 +129281,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -129292,11 +129292,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -129307,11 +129307,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -129322,11 +129322,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -129337,26 +129337,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -129367,26 +129367,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -129397,26 +129397,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -129427,15 +129427,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -129445,8 +129445,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129460,8 +129460,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129475,8 +129475,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129490,8 +129490,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129505,8 +129505,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129520,8 +129520,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129535,8 +129535,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -129949,9 +129949,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -129961,7 +129961,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -129986,8 +129986,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -129997,11 +129997,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130012,11 +130012,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130027,11 +130027,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130042,26 +130042,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130072,26 +130072,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130102,26 +130102,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130132,15 +130132,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -130150,8 +130150,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130165,8 +130165,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130180,8 +130180,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130195,8 +130195,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130210,8 +130210,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130225,8 +130225,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130240,8 +130240,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130654,9 +130654,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -130666,7 +130666,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -130691,8 +130691,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -130702,11 +130702,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130717,11 +130717,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130732,11 +130732,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130747,26 +130747,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130777,26 +130777,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130807,26 +130807,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -130837,15 +130837,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -130855,8 +130855,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130870,8 +130870,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130885,8 +130885,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130900,8 +130900,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130915,8 +130915,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130930,8 +130930,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -130945,8 +130945,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131359,9 +131359,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -131371,7 +131371,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -131396,8 +131396,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -131407,11 +131407,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -131422,11 +131422,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -131437,11 +131437,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -131452,26 +131452,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -131482,26 +131482,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -131512,26 +131512,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -131542,15 +131542,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -131560,8 +131560,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131575,8 +131575,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131590,8 +131590,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131605,8 +131605,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131620,8 +131620,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131635,8 +131635,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -131650,8 +131650,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132064,9 +132064,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -132076,7 +132076,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -132101,8 +132101,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -132112,11 +132112,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132127,11 +132127,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132142,11 +132142,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132157,26 +132157,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132187,26 +132187,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132217,26 +132217,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132247,15 +132247,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -132265,8 +132265,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132280,8 +132280,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132295,8 +132295,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132310,8 +132310,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132325,8 +132325,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132340,8 +132340,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132355,8 +132355,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132769,9 +132769,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -132781,7 +132781,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -132806,8 +132806,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -132817,11 +132817,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132832,11 +132832,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132847,11 +132847,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132862,26 +132862,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132892,26 +132892,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132922,26 +132922,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -132952,15 +132952,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -132970,8 +132970,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -132985,8 +132985,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133000,8 +133000,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133015,8 +133015,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133030,8 +133030,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133045,8 +133045,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133060,8 +133060,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133474,9 +133474,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -133486,7 +133486,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -133511,8 +133511,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -133522,11 +133522,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -133537,11 +133537,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -133552,11 +133552,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -133567,26 +133567,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -133597,26 +133597,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -133627,26 +133627,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -133657,15 +133657,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -133675,8 +133675,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133690,8 +133690,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133705,8 +133705,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133720,8 +133720,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133735,8 +133735,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133750,8 +133750,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -133765,8 +133765,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134179,9 +134179,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -134191,7 +134191,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -134216,8 +134216,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -134227,11 +134227,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134242,11 +134242,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134257,11 +134257,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134272,26 +134272,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134302,26 +134302,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134332,26 +134332,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134362,15 +134362,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -134380,8 +134380,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134395,8 +134395,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134410,8 +134410,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134425,8 +134425,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134440,8 +134440,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134455,8 +134455,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134470,8 +134470,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -134884,9 +134884,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -134896,7 +134896,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -134921,8 +134921,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -134932,11 +134932,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134947,11 +134947,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134962,11 +134962,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -134977,26 +134977,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135007,26 +135007,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135037,26 +135037,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135067,15 +135067,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -135085,8 +135085,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135100,8 +135100,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135115,8 +135115,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135130,8 +135130,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135145,8 +135145,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135160,8 +135160,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135175,8 +135175,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135589,9 +135589,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -135601,7 +135601,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -135626,8 +135626,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -135637,11 +135637,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135652,11 +135652,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135667,11 +135667,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135682,26 +135682,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135712,26 +135712,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135742,26 +135742,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -135772,15 +135772,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -135790,8 +135790,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135805,8 +135805,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135820,8 +135820,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135835,8 +135835,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135850,8 +135850,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135865,8 +135865,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -135880,8 +135880,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136294,9 +136294,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -136306,7 +136306,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -136331,8 +136331,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -136342,11 +136342,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -136357,11 +136357,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -136372,11 +136372,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -136387,26 +136387,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -136417,26 +136417,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -136447,26 +136447,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -136477,15 +136477,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -136495,8 +136495,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136510,8 +136510,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136525,8 +136525,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136540,8 +136540,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136555,8 +136555,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136570,8 +136570,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136585,8 +136585,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -136999,9 +136999,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -137011,7 +137011,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -137036,8 +137036,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -137047,11 +137047,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137062,11 +137062,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137077,11 +137077,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137092,26 +137092,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137122,26 +137122,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137152,26 +137152,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137182,15 +137182,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -137200,8 +137200,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137215,8 +137215,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137230,8 +137230,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137245,8 +137245,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137260,8 +137260,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137275,8 +137275,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137290,8 +137290,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137704,9 +137704,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -137716,7 +137716,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -137741,8 +137741,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -137752,11 +137752,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137767,11 +137767,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137782,11 +137782,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137797,26 +137797,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137827,26 +137827,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137857,26 +137857,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -137887,15 +137887,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -137905,8 +137905,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137920,8 +137920,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137935,8 +137935,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137950,8 +137950,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137965,8 +137965,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137980,8 +137980,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -137995,8 +137995,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138409,9 +138409,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -138421,7 +138421,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -138446,8 +138446,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -138457,11 +138457,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -138472,11 +138472,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -138487,11 +138487,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -138502,26 +138502,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -138532,26 +138532,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -138562,26 +138562,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -138592,15 +138592,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -138610,8 +138610,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138625,8 +138625,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138640,8 +138640,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138655,8 +138655,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138670,8 +138670,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138685,8 +138685,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -138700,8 +138700,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139114,9 +139114,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -139126,7 +139126,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -139151,8 +139151,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -139162,11 +139162,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139177,11 +139177,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139192,11 +139192,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139207,26 +139207,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139237,26 +139237,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139267,26 +139267,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139297,15 +139297,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -139315,8 +139315,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139330,8 +139330,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139345,8 +139345,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139360,8 +139360,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139375,8 +139375,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139390,8 +139390,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139405,8 +139405,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -139819,9 +139819,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -139831,7 +139831,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -139856,8 +139856,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -139867,11 +139867,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139882,11 +139882,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139897,11 +139897,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139912,26 +139912,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139942,26 +139942,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -139972,26 +139972,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140002,15 +140002,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -140020,8 +140020,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140035,8 +140035,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140050,8 +140050,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140065,8 +140065,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140080,8 +140080,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140095,8 +140095,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140110,8 +140110,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140524,9 +140524,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -140536,7 +140536,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -140561,8 +140561,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -140572,11 +140572,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140587,11 +140587,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140602,11 +140602,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140617,26 +140617,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140647,26 +140647,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140677,26 +140677,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -140707,15 +140707,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -140725,8 +140725,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140740,8 +140740,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140755,8 +140755,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140770,8 +140770,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140785,8 +140785,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140800,8 +140800,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -140815,8 +140815,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141229,9 +141229,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -141241,7 +141241,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -141266,8 +141266,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -141277,11 +141277,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141292,11 +141292,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141307,11 +141307,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141322,26 +141322,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141352,26 +141352,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141382,26 +141382,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141412,15 +141412,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -141430,8 +141430,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141445,8 +141445,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141460,8 +141460,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141475,8 +141475,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141490,8 +141490,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141505,8 +141505,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141520,8 +141520,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -141934,9 +141934,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -141946,7 +141946,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -141971,8 +141971,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -141982,11 +141982,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -141997,11 +141997,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142012,11 +142012,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142027,26 +142027,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142057,26 +142057,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142087,26 +142087,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142117,15 +142117,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -142135,8 +142135,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142150,8 +142150,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142165,8 +142165,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142180,8 +142180,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142195,8 +142195,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142210,8 +142210,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142225,8 +142225,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142639,9 +142639,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -142651,7 +142651,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -142676,8 +142676,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -142687,11 +142687,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142702,11 +142702,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142717,11 +142717,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142732,26 +142732,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142762,26 +142762,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142792,26 +142792,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -142822,15 +142822,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -142840,8 +142840,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142855,8 +142855,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142870,8 +142870,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142885,8 +142885,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142900,8 +142900,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142915,8 +142915,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -142930,8 +142930,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143344,9 +143344,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -143356,7 +143356,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -143381,8 +143381,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -143392,11 +143392,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -143407,11 +143407,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -143422,11 +143422,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -143437,26 +143437,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -143467,26 +143467,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -143497,26 +143497,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -143527,15 +143527,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -143545,8 +143545,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143560,8 +143560,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143575,8 +143575,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143590,8 +143590,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143605,8 +143605,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143620,8 +143620,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -143635,8 +143635,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144049,9 +144049,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -144061,7 +144061,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -144086,8 +144086,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -144097,11 +144097,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144112,11 +144112,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144127,11 +144127,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144142,26 +144142,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144172,26 +144172,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144202,26 +144202,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144232,15 +144232,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -144250,8 +144250,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144265,8 +144265,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144280,8 +144280,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144295,8 +144295,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144310,8 +144310,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144325,8 +144325,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144340,8 +144340,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144754,9 +144754,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -144766,7 +144766,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -144791,8 +144791,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -144802,11 +144802,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144817,11 +144817,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144832,11 +144832,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144847,26 +144847,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144877,26 +144877,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144907,26 +144907,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -144937,15 +144937,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -144955,8 +144955,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144970,8 +144970,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -144985,8 +144985,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145000,8 +145000,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145015,8 +145015,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145030,8 +145030,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145045,8 +145045,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145459,9 +145459,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -145471,7 +145471,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -145496,8 +145496,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -145507,11 +145507,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -145522,11 +145522,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -145537,11 +145537,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -145552,26 +145552,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -145582,26 +145582,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -145612,26 +145612,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -145642,15 +145642,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -145660,8 +145660,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145675,8 +145675,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145690,8 +145690,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145705,8 +145705,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145720,8 +145720,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145735,8 +145735,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -145750,8 +145750,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146164,9 +146164,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -146176,7 +146176,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -146201,8 +146201,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -146212,11 +146212,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146227,11 +146227,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed convergence comparison",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146242,11 +146242,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146257,26 +146257,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146287,26 +146287,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146317,26 +146317,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146347,15 +146347,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -146365,8 +146365,8 @@
               "representation": "lower-PTX and disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146380,8 +146380,8 @@
               "representation": "disallowed-target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146395,8 +146395,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146410,8 +146410,8 @@
               "representation": "missing exact-target suffix boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146425,8 +146425,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146440,8 +146440,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146455,8 +146455,8 @@
               "representation": "lower-PTX target boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146850,9 +146850,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -146862,7 +146862,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -146875,8 +146875,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -146886,11 +146886,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and predicate audit",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146901,11 +146901,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k32 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -146916,15 +146916,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k32 F16 ptxas-safe floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -146934,8 +146934,8 @@
               "representation": "k32 F16 ptxas-safe lower boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -146949,8 +146949,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -147212,9 +147212,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -147224,7 +147224,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -147237,8 +147237,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -147248,11 +147248,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and predicate audit",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -147263,11 +147263,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k32 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -147278,15 +147278,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k32 F16 ptxas-safe floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -147296,8 +147296,8 @@
               "representation": "k32 F16 ptxas-safe lower boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -147311,8 +147311,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -147574,9 +147574,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -147586,7 +147586,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -147599,8 +147599,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -147610,11 +147610,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and predicate audit",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -147625,11 +147625,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k32 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -147640,15 +147640,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k32 F16 ptxas-safe floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -147658,8 +147658,8 @@
               "representation": "k32 F16 ptxas-safe lower boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -147673,8 +147673,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -147936,9 +147936,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -147948,7 +147948,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -147961,8 +147961,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -147972,11 +147972,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "typed LLVM declarations and predicate audit",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -147987,11 +147987,11 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "convergent k32 F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -148002,15 +148002,15 @@
                 "sm_89",
                 "ptx87"
               ],
-              "representation": "k32 F16 ptxas-safe floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -148020,8 +148020,8 @@
               "representation": "k32 F16 ptxas-safe lower boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -148035,8 +148035,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -148306,9 +148306,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -148318,7 +148318,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -148331,8 +148331,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -148342,11 +148342,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all four k32/F32 declarations without convergent.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -148357,11 +148357,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "convergent k32 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -148372,15 +148372,15 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "k32 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -148390,8 +148390,8 @@
               "representation": "k32 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -148405,8 +148405,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -148676,9 +148676,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -148688,7 +148688,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -148701,8 +148701,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -148712,11 +148712,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all four k32/F32 declarations without convergent.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -148727,11 +148727,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "convergent k32 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -148742,15 +148742,15 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "k32 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -148760,8 +148760,8 @@
               "representation": "k32 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -148775,8 +148775,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -149046,9 +149046,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -149058,7 +149058,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -149071,8 +149071,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -149082,11 +149082,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all four k32/F32 declarations without convergent.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149097,11 +149097,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "convergent k32 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149112,15 +149112,15 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "k32 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -149130,8 +149130,8 @@
               "representation": "k32 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -149145,8 +149145,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -149416,9 +149416,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -149428,7 +149428,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -149441,8 +149441,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -149452,11 +149452,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "typed LLVM declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized all four k32/F32 declarations without convergent.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149467,11 +149467,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "convergent k32 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149482,15 +149482,15 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "k32 F32 safe PTX floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -149500,8 +149500,8 @@
               "representation": "k32 F32 lower-PTX boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -149515,8 +149515,8 @@
               "representation": "lower-SM boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -149815,9 +149815,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -149827,7 +149827,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -149852,8 +149852,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -149863,11 +149863,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149878,26 +149878,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149908,26 +149908,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149938,26 +149938,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -149968,15 +149968,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -150405,9 +150405,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -150417,7 +150417,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -150442,8 +150442,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -150453,11 +150453,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -150468,26 +150468,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -150498,26 +150498,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -150528,26 +150528,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -150558,15 +150558,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -150995,9 +150995,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -151007,7 +151007,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -151032,8 +151032,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -151043,11 +151043,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151058,26 +151058,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151088,26 +151088,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151118,26 +151118,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151148,15 +151148,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -151585,9 +151585,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -151597,7 +151597,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -151622,8 +151622,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -151633,11 +151633,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151648,26 +151648,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151678,26 +151678,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151708,26 +151708,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -151738,15 +151738,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -152175,9 +152175,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -152187,7 +152187,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -152212,8 +152212,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -152223,11 +152223,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152238,26 +152238,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152268,26 +152268,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152298,26 +152298,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152328,15 +152328,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -152765,9 +152765,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -152777,7 +152777,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -152802,8 +152802,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -152813,11 +152813,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152828,26 +152828,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152858,26 +152858,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152888,26 +152888,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -152918,15 +152918,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -153355,9 +153355,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -153367,7 +153367,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -153392,8 +153392,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -153403,11 +153403,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -153418,26 +153418,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -153448,26 +153448,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -153478,26 +153478,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -153508,15 +153508,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -153945,9 +153945,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -153957,7 +153957,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -153982,8 +153982,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -153993,11 +153993,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154008,26 +154008,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154038,26 +154038,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154068,26 +154068,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154098,15 +154098,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -154535,9 +154535,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -154547,7 +154547,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -154572,8 +154572,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -154583,11 +154583,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154598,26 +154598,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154628,26 +154628,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154658,26 +154658,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -154688,15 +154688,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -155125,9 +155125,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -155137,7 +155137,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -155162,8 +155162,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -155173,11 +155173,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155188,26 +155188,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155218,26 +155218,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155248,26 +155248,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155278,15 +155278,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -155715,9 +155715,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -155727,7 +155727,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -155752,8 +155752,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -155763,11 +155763,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155778,26 +155778,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155808,26 +155808,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155838,26 +155838,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -155868,15 +155868,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -156305,9 +156305,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -156317,7 +156317,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -156342,8 +156342,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -156353,11 +156353,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -156368,26 +156368,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -156398,26 +156398,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -156428,26 +156428,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -156458,15 +156458,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -156895,9 +156895,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -156907,7 +156907,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -156932,8 +156932,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -156943,11 +156943,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -156958,26 +156958,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -156988,26 +156988,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -157018,26 +157018,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -157048,15 +157048,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -157485,9 +157485,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -157497,7 +157497,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -157522,8 +157522,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -157533,11 +157533,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -157548,26 +157548,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -157578,26 +157578,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -157608,26 +157608,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -157638,15 +157638,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -158075,9 +158075,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -158087,7 +158087,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -158112,8 +158112,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -158123,11 +158123,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158138,26 +158138,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158168,26 +158168,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158198,26 +158198,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158228,15 +158228,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -158665,9 +158665,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -158677,7 +158677,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -158702,8 +158702,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -158713,11 +158713,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158728,26 +158728,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158758,26 +158758,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158788,26 +158788,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -158818,15 +158818,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -159255,9 +159255,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -159267,7 +159267,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -159292,8 +159292,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -159303,11 +159303,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159318,26 +159318,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159348,26 +159348,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159378,26 +159378,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159408,15 +159408,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -159845,9 +159845,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -159857,7 +159857,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -159882,8 +159882,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -159893,11 +159893,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159908,26 +159908,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159938,26 +159938,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159968,26 +159968,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -159998,15 +159998,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -160435,9 +160435,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -160447,7 +160447,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -160472,8 +160472,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -160483,11 +160483,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -160498,26 +160498,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -160528,26 +160528,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -160558,26 +160558,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -160588,15 +160588,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -161025,9 +161025,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -161037,7 +161037,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -161062,8 +161062,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -161073,11 +161073,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161088,26 +161088,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161118,26 +161118,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161148,26 +161148,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161178,15 +161178,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -161615,9 +161615,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -161627,7 +161627,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -161652,8 +161652,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -161663,11 +161663,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161678,26 +161678,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161708,26 +161708,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161738,26 +161738,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -161768,15 +161768,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -162205,9 +162205,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -162217,7 +162217,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -162242,8 +162242,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -162253,11 +162253,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162268,26 +162268,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162298,26 +162298,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162328,26 +162328,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162358,15 +162358,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -162795,9 +162795,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -162807,7 +162807,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -162832,8 +162832,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -162843,11 +162843,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162858,26 +162858,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162888,26 +162888,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162918,26 +162918,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -162948,15 +162948,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -163385,9 +163385,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -163397,7 +163397,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -163422,8 +163422,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -163433,11 +163433,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -163448,26 +163448,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -163478,26 +163478,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -163508,26 +163508,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -163538,15 +163538,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -163975,9 +163975,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -163987,7 +163987,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -164012,8 +164012,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -164023,11 +164023,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -164038,26 +164038,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -164068,26 +164068,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -164098,26 +164098,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -164128,15 +164128,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact dense kind::mxf8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -164530,9 +164530,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -164542,7 +164542,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -164555,8 +164555,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -164566,11 +164566,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 s4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47aa6dd3e24383098f128a64c87c47ab4797f5e033e3fc4bb0a670d2503013a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -164581,15 +164581,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3aa0206bc409a773e92374c337ae29700fa54735792eeaafe1d8f87fcb9c523.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -164803,9 +164803,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -164815,7 +164815,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -164828,8 +164828,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -164839,11 +164839,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 s4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d9b3d3e50e66a2281bd4a40ade4118f0afb4d7789eef88afc79e9498ac5a3ea2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -164854,15 +164854,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ca1b55fd6cb49b5380574ff42ec7d55b6dbf557c08ff72873a5da7eecd90838.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -165077,9 +165077,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -165089,7 +165089,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -165102,8 +165102,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -165113,11 +165113,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 s4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab9e908618637e0d0699f72f1a3a80b33f56c41e57a5c802af03819ef2c5ef14.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -165128,15 +165128,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7567be8747a7b101aa8df77028eef647926a3d0dea2f0fcd83d3ebe953a622e2.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -165350,9 +165350,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -165362,7 +165362,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -165375,8 +165375,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -165386,11 +165386,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 s4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 81ae38ebf4bd8a59ff18a33203383a22cace76e383f473b31a6debf0f9ee947f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -165401,15 +165401,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 973c539e9cb7d01605f08ede31f1bcf9fb12155ae80857ea6da2f5bf085509f6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -165630,9 +165630,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -165642,7 +165642,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -165655,8 +165655,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -165666,11 +165666,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for signed INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact signed INT8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6524c9c1c5d31e51a7bc4f69ed88bb6f64bc9718c4cd4e6dbdf38d71cceb7406.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -165681,15 +165681,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f67ea6f8fc76d2142c8297e2fbef29257d5a89eba91380d4ebbbf47cdce91489.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -165909,9 +165909,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -165921,7 +165921,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -165934,8 +165934,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -165945,11 +165945,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k32 signed INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite s8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c512bacbd7f6da60ef9eda11b065823f600b9f0409a0ca562d8b2ae0e2c2df48.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -165960,15 +165960,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57e571ed96c84be6aaa7c9428a16c4c002f791b55429f427820fcc3d6fbb177c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -166189,9 +166189,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -166201,7 +166201,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -166214,8 +166214,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -166225,11 +166225,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping s8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 77110080515c64834cb555469a0ea69ed12c499b30be821f8832fdcfac0f73c5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -166240,15 +166240,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7700bbce0406dc6977a0af2702a6679b1a41e88c867eed3903621b9d14beef53.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -166468,9 +166468,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -166480,7 +166480,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -166493,8 +166493,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -166504,11 +166504,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k32 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite s8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16e7d991475a0e8577e129c8337f9eb2ba08ec792a3500cd9890a5be58311b2b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -166519,15 +166519,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 74351d962b7eb0ac0ca27e606b8d6e75468b39752b667b28e6277475de18bb29.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -166742,9 +166742,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -166754,7 +166754,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -166767,8 +166767,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -166778,11 +166778,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 u4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9cd0a0a84b94f37267419e33805b31860d5685fcd2cf4564783e28343a972748.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -166793,15 +166793,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f2bac6e9f7ce9e6cb01846a046917c140500e7e876942100e7d3ce977cbfd90d.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -167015,9 +167015,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -167027,7 +167027,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -167040,8 +167040,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -167051,11 +167051,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 u4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a51cb813a45b9cd0c0e9dc7038bec77a14194732f4b180ec96eecd3958d6a169.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -167066,15 +167066,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b2c092a2e9c78fd3580523dc6cc875e8903497e790eee28cfa15324c03685a7.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -167288,9 +167288,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -167300,7 +167300,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -167313,8 +167313,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -167324,11 +167324,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 u4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ce1cd3a7682dc726d32cf180d1b39bdf9ac3aea8f12aa2a6c6423b7f521a15ab.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -167339,15 +167339,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a5085e6a26e349f53eba46f4e630ecd2142786cbed096ed927acc94cc9c5d910.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -167562,9 +167562,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -167574,7 +167574,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -167587,8 +167587,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -167598,11 +167598,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 u4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 023ff2fe53a4cd8bb534791bde0e7384b5820f1fb7eb9cb1bed3627b8bed4c7e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -167613,15 +167613,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fedf24465346088c564e7971f6b7c29e92fceb052adb100fafb91b5e5b3b6bee.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -167842,9 +167842,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -167854,7 +167854,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -167867,8 +167867,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -167878,11 +167878,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 unsigned INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping u8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67ac1fcd13f430b3a14dc6d8467cc9beec1a0b175c1e785fce2a9a91545d379d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -167893,15 +167893,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ba09c1ddf924a125f7171df66dec143a9c4d598837bdaed8de694fc7b52b4b8.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -168121,9 +168121,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -168133,7 +168133,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -168146,8 +168146,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -168157,11 +168157,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k32 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping u8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ec734a83add935db593a1f90090b395a74af19c9c84661e61a80d46ba07966ca.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -168172,15 +168172,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13d684b6400fcb97db851b1a163797e7b6cabd462f847f3f54b3342a0d49697c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -168400,9 +168400,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -168412,7 +168412,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -168425,8 +168425,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -168436,11 +168436,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k32 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite u8-by-s8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f2c069f3d47afb6653954d8f58fb146b9e927136d703104a92a2b8312c874d45.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -168451,15 +168451,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fffbf71c999883fa695a7300a982460fc7e6425a322521cc9e98523dcfdc3fc2.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -168680,9 +168680,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -168692,7 +168692,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -168705,8 +168705,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -168716,11 +168716,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for saturating m16n8k32 unsigned INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite u8-by-u8 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eaabf735e8ebc1b813e38c3ec0d3c20f167786603c98ce8ebc1867abdf0e5d52.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -168731,15 +168731,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 deefb4ba049125c5b91b39ac823909f05e24c773ceac2520b10210d11e02c17b.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -168954,9 +168954,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -168966,7 +168966,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -168979,8 +168979,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -168993,8 +168993,8 @@
               "representation": "m16n8k4 TF32 lower-hardware boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the form below sm_80; the diagnostic SHA-256 is 73ef62d63bacc273dbfc152d2cb3acb93bfe3c744b3003e49cf66cd8a47e0b5d.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the form below sm_80; the diagnostic SHA-256 is 73ef62d63bacc273dbfc152d2cb3acb93bfe3c744b3003e49cf66cd8a47e0b5d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -169005,11 +169005,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five typed LLVM declarations and duplicate-call probes",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -169020,11 +169020,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five convergent side-effecting register-only inline PTX calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -169035,15 +169035,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "aggregate inline-PTX MMA fixture at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -169278,9 +169278,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -169290,7 +169290,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -169303,8 +169303,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -169314,11 +169314,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 s4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5ff5fcfbf7af82721582a788d144fe6745bd1ea783aef2ff9102a328612ce3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -169329,15 +169329,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95520ee8bd9137e246a8f6b10d90bf414fd6640b6941209df067082273bab6a4.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -169557,9 +169557,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -169569,7 +169569,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -169582,8 +169582,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -169593,11 +169593,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 s4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b42070729512dc4a34d959933cb2ed579c549ca2c4a34ba53634c793c2104c5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -169608,15 +169608,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 052fb66d0981ed2361e6cde56d62007088995b7ba713d6da1691e0012e7f4923.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -169837,9 +169837,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -169849,7 +169849,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -169862,8 +169862,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -169873,11 +169873,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 s4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b6f9f1214b02bafdfe07cca12072fdbfab865362fd3b295e688a738be8d0df.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -169888,15 +169888,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 814a279d3913788ca9cc088325b84198c97e9693aaa13778c28f0f654b834e38.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -170116,9 +170116,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -170128,7 +170128,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -170141,8 +170141,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -170152,11 +170152,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 s4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6d117f7dc3771deff164ddacd83b64a80d1b9c3f33543c2323394f860296b3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -170167,15 +170167,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 232ee3454e7542b49e296aafcd4c9f38235fe2141fb224ca3f329aec36d456b3.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -170396,9 +170396,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -170408,7 +170408,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -170421,8 +170421,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -170432,11 +170432,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 u4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 48af8df444ba8fcff6c181e3862bbcfabfeb9817b970ddb66ea95df74814252e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -170447,15 +170447,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7e02b52b0a757ec780b5e93f59ff1abdea9770a82d4f0f4d1384035db9f803b.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -170675,9 +170675,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -170687,7 +170687,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -170700,8 +170700,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -170711,11 +170711,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 u4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c9140abf7847a4b891673bba722aaef6a63ce9520ae3764efd411de6d636208f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -170726,15 +170726,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3824b5037253381a849ded485ed000a3c459f68a94dcd43b809a72958bf618aa.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -170954,9 +170954,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -170966,7 +170966,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -170979,8 +170979,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -170990,11 +170990,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 u4-by-s4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa313e124cffc714546231b65e670c63303446e7edb95dce4af94ca57bf258a3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171005,15 +171005,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ed63a17572d63f68c5f40f43cf5920d3432723ec6b946bd1689e3c44d3fe4f9.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -171234,9 +171234,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -171246,7 +171246,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -171259,8 +171259,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -171270,11 +171270,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m16n8k64 u4-by-u4 INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 59332065be960c742528cb03bad200c2c5a5ad472285420dd46116d3d9da8d88.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171285,15 +171285,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2b6ed03aae00d63ec51d849f1382eb95b8580686067589bea8eaa9433e3d8be1.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -171500,9 +171500,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -171512,7 +171512,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -171525,8 +171525,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -171536,11 +171536,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five typed LLVM declarations and duplicate-call probes",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171551,11 +171551,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five convergent side-effecting register-only inline PTX calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171566,15 +171566,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "aggregate inline-PTX MMA fixture at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -171803,9 +171803,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -171815,7 +171815,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -171828,8 +171828,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -171842,8 +171842,8 @@
               "representation": "m16n8k8 BF16 imported-predicate boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM selects this form at its imported sm_75/PTX 6.5 predicate, but ptxas rejects it even at PTX 7.0 because this BF16 m16n8k8 form requires sm_80. The PTX 7.0 diagnostic SHA-256 is 69f7da673aef781e3eb0c06d602ccab1fdc5d95e4b0d645d6e118fd55d0747c8.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM selects this form at its imported sm_75/PTX 6.5 predicate, but ptxas rejects it even at PTX 7.0 because this BF16 m16n8k8 form requires sm_80. The PTX 7.0 diagnostic SHA-256 is 69f7da673aef781e3eb0c06d602ccab1fdc5d95e4b0d645d6e118fd55d0747c8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171854,11 +171854,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five typed LLVM declarations and duplicate-call probes",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171869,11 +171869,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five convergent side-effecting register-only inline PTX calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -171884,15 +171884,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "aggregate inline-PTX MMA fixture at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -172121,9 +172121,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -172133,7 +172133,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -172146,8 +172146,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -172157,11 +172157,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five typed LLVM declarations and duplicate-call probes",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -172172,11 +172172,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "five convergent side-effecting register-only inline PTX calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -172187,15 +172187,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "aggregate inline-PTX MMA fixture at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29.",
+              "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -172430,9 +172430,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -172442,7 +172442,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -172455,8 +172455,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -172466,11 +172466,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for TF32 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact TF32 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 229d640dcb7295b409e4d690f6094a704af0b77e69e834abb5c2c9d26ac5daa0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -172481,15 +172481,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7543fed70ec348791089d56f1eee6d51e13501a1a40092f118bec2c99b734d3c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -172694,9 +172694,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -172706,7 +172706,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -172719,8 +172719,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -172730,11 +172730,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k128 B1 AND-popcount MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6cf34f19543565ab3b0a268c1b38020771261bc358df589709391ec5e99ad3f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -172745,15 +172745,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for m8n8k128 B1 AND-popcount MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93107316bf0cf5ef84b40e52ec5ef683b2781a3e3a0e75737e4892184d0c6457.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -172959,9 +172959,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -172971,7 +172971,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -172984,8 +172984,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -172995,11 +172995,11 @@
                 "sm_75",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k128 B1 XOR-popcount MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 666182ec743c9ffe0422975ab4ab4faf2e3ddcff656913f34303f15200fa3700.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -173010,15 +173010,15 @@
                 "sm_75",
                 "ptx70"
               ],
-              "representation": "exact-floor PTX for m8n8k128 B1 XOR-popcount MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_75 floors.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 452782b4303bcdb3357c76efeae34f913d781f5673cd5a97eace443e1dc28b77.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -173224,9 +173224,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -173236,7 +173236,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -173249,8 +173249,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -173260,11 +173260,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k16 signed INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping s8-by-s8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5c2da0b56637465b6318014f3bb75f68d843c14ef827e70a7ef937625ad5e2c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -173275,15 +173275,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df22078cac5b20809b7d5a9eea6a29d6cfe354500d77fa360d86a0b4a1023852.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -173487,9 +173487,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -173499,7 +173499,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -173512,8 +173512,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -173523,11 +173523,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k16 signed INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite s8-by-s8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 270630238dd5c5f9cf517a25b950096e73e12e4479f81ac8c1cd3283017390d0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -173538,15 +173538,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1bf5493b1e849d63814126e394e6f0572529d702a2433d0de1ff47b7b4ac0650.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -173751,9 +173751,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -173763,7 +173763,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -173776,8 +173776,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -173787,11 +173787,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping s8-by-u8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab1d12b93c4b140169d3169350a16500008aa4163f09614a631b7e1dc4fdbcb1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -173802,15 +173802,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a423b06a09e13590b1d54c6f8d8be51d1e394145354b004a635293e1d76ac83c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -174014,9 +174014,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -174026,7 +174026,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -174039,8 +174039,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -174050,11 +174050,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite s8-by-u8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d10485a6ecf25b1ef218ba5ca3641e58f8066afb750fe46d83566b3fe2b4f072.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -174065,15 +174065,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f5aaa3aac438dbac75e659a7a874e86e99e3f0114d38a53f3d889f2063ea0b6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -174278,9 +174278,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -174290,7 +174290,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -174303,8 +174303,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -174314,11 +174314,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k16 unsigned INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping u8-by-u8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f6d577d18158476e7ab49d17a2ad0f47da8b6ae752a6ed080d20e1380d52bee.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -174329,15 +174329,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1955df76beb6ee81ed7ea879c3200f4f6bb7e1966deb92e945b3f325841dc5cd.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -174541,9 +174541,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -174553,7 +174553,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -174566,8 +174566,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -174577,11 +174577,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact wrapping u8-by-s8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f75f6988bc1dda288d7c571d72462c4298a17b37aaa830a033de583852ac772.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -174592,15 +174592,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4dff1215835b5b26444fd0e25f94789efea21a227da12d555c889c796c0ea93f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -174804,9 +174804,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -174816,7 +174816,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -174829,8 +174829,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -174840,11 +174840,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k16 mixed-sign INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite u8-by-s8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93bd9995092da3436c05eb1720405ceeea64728cf2eb687f907c7b1c820d0148.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -174855,15 +174855,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67dbf19d3684c67bd2b2ffface224d08f9e24a216e617ebf5307e530bc492d97.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -175068,9 +175068,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -175080,7 +175080,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -175093,8 +175093,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -175104,11 +175104,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k16 unsigned INT8 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact satfinite u8-by-u8 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51db7a6fe697ef26e445bac4abdaa4d5e45e887e8aa5a366f608654d3466aab0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -175119,15 +175119,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06aafc80a8ca8190b809819eaac9c4776807f6380a05ed5a02ce479898506506.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -175332,9 +175332,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -175344,7 +175344,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -175357,8 +175357,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -175368,11 +175368,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k32 signed INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping s4-by-s4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 242a7d2104a977a3c344c947eb5c03d10a18711cb776b10a27c2519d3760eeb0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -175383,15 +175383,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c5b6ee75c034c9af33c18ab6c91f725b42d2b54c8bf132ec1b89c115cb9a3548.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -175595,9 +175595,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -175607,7 +175607,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -175620,8 +175620,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -175631,11 +175631,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k32 signed INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite s4-by-s4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5fb1110c752bf4fe08f37bd8b7f8345d676ca3eea504e71bed250fb3c7497ce.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -175646,15 +175646,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e9c1003612b5e1b0441919a9d51eeadbd9a9c26f365be978cf5d279b0b0d1fc.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -175859,9 +175859,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -175871,7 +175871,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -175884,8 +175884,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -175895,11 +175895,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k32 mixed-sign INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping s4-by-u4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57e936cd723c8522e8d7ad7c20eec7f1efe473d9dde5381bc74874bbd737acc1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -175910,15 +175910,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 540c2a2b1c2c2945022dad6f87d3cdf948ade2eab7a1aa2149e47b320adf4e48.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -176122,9 +176122,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -176134,7 +176134,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -176147,8 +176147,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -176158,11 +176158,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k32 mixed-sign INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite s4-by-u4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1a0ad568976eb0b009eeb8941f96e74169d76cd8f5e486c761848b16a6a78f68.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -176173,15 +176173,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ee342c6e9f3f578c32c6af5dd1c2668cf556669258abfcce6f85298f77a2fe90.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -176386,9 +176386,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -176398,7 +176398,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -176411,8 +176411,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -176422,11 +176422,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k32 unsigned INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping u4-by-u4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16ea31bd434789327320d0683d1cc83a11058292365655503c77c2143139e1e4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -176437,15 +176437,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75676b7cb4e4fb123537b9f8f1cc40ccb04e1db0271ab3eb760961f4ea73ea37.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -176649,9 +176649,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -176661,7 +176661,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -176674,8 +176674,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -176685,11 +176685,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for m8n8k32 mixed-sign INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping u4-by-s4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a777e8a890e81f7bc799e08ffa31129ff28d46f13cd9bbb9467f7d3e09e294b6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -176700,15 +176700,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 87501053789841d181fd1aae8dac5b4dc516639d99983f1867fd6f4df9be4836.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -176912,9 +176912,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -176924,7 +176924,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -176937,8 +176937,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -176948,11 +176948,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k32 mixed-sign INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite u4-by-s4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d3a2ccb17359a5a84b5f2ffbbbf21bdbf3e862d6a2a6e007cba7442aa65144a9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -176963,15 +176963,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 787d75bc93a0a84ac830fa9dd7472adc96e9b78362d7e6ce1b107301d5c6d0a6.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -177176,9 +177176,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -177188,7 +177188,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.5",
             "hardware": {
@@ -177201,8 +177201,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -177212,11 +177212,11 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "convergent register-only inline PTX for saturating m8n8k32 unsigned INT4 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite u4-by-u4 instruction in C, A, B API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b77eb2760494d673de880beb0e9da2110a690897b8968ba0b33d57e0f7b635f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -177227,15 +177227,15 @@
                 "sm_75",
                 "ptx65"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 666fe69ffd45aab6a3f5bfc5183c41460a862957228ada97aee5dece4d4b92a4.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -177440,9 +177440,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -177452,7 +177452,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -177465,8 +177465,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -177476,11 +177476,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "convergent register-only inline PTX for f64 MMA",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The generated probe emitted the exact f64 MMA instruction with C, A, B operands kept in public API order.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 455ca35747a458536fae20c94d1749e6b7c07dc9b482fb469192376c9e10a1a8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -177491,15 +177491,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f7bf4efce75e805867c6ef541acdd239b2df1f6819376e1d08b24cf76398b507.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -177731,9 +177731,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -177743,7 +177743,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -177756,8 +177756,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -177767,11 +177767,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 wrapping s4-by-s4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f178dc6498dfe1335f303815c77a74f6c64eef9102c2ae74312466d8ac126ab0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -177782,15 +177782,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e32824932666209d5959192f45f412c7f087fd40cd3b8a97e82e9004721855c5.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -178046,9 +178046,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -178058,7 +178058,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -178071,8 +178071,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -178082,11 +178082,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 saturating s4-by-s4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f05e41f54c80df9399f1535e4630d9ac644c42c0c8bb66c19902effd9fe96f3b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -178097,15 +178097,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2733247c94953620e2bdfdda7e22bb9f6a0f731285caa41a50d70a50d63c92f3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -178362,9 +178362,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -178374,7 +178374,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -178387,8 +178387,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -178398,11 +178398,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 wrapping s4-by-u4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 751f4ccde1a953061a6de9f03fc03e29bcab215d3e0dfaf5cadec2fdab2ffda9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -178413,15 +178413,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4b7f6a5ffd37de7a86d2c35f122c69356b19e3ab05c2eaf64e726dd677fffd2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -178677,9 +178677,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -178689,7 +178689,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -178702,8 +178702,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -178713,11 +178713,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 saturating s4-by-u4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e91aa50a469e4ae22cabf614f0c86c11a381dd99b7d57a4f376c7f86b815819.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -178728,15 +178728,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c0e9109eafe381fc83456148a9518994753df9868388321037f0cbe11c64a260.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -178993,9 +178993,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -179005,7 +179005,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -179018,8 +179018,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -179029,11 +179029,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 wrapping u4-by-u4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa743b0396151d7184adfa37f3ff722cd693b7bdb74f433a522498874947fbfb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -179044,15 +179044,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 56253d8e8cd56f03dadc8369add61af230eac577a3258c42a28dcf4f2ede4f3d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -179308,9 +179308,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -179320,7 +179320,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -179333,8 +179333,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -179344,11 +179344,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 wrapping u4-by-s4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6c8d0ebc3993fe3649feb8dc574ca4aa6d9bbb8a54b57bab64439c72b3183f91.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -179359,15 +179359,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e93fae493c1bff6c15d4b462fe8d3e9926f99da8c339f07247d5633515195e36.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -179623,9 +179623,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -179635,7 +179635,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -179648,8 +179648,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -179659,11 +179659,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 saturating u4-by-s4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75d63c6f58dbf64b7ff7c0bff1b89fd0c2c7776ecc86997ccfcd3184130f610f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -179674,15 +179674,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3f4ba91d089260787eb84c1e93a4b9ad6ea30660ea314e755f025574e205657.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -179939,9 +179939,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -179951,7 +179951,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -179964,8 +179964,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -179975,11 +179975,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k128 saturating u4-by-u4 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f13fef29cc4b67af7e5a181bef0b806732d7eea0ec0f5176e91581b982f20511.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -179990,15 +179990,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0557fe188cb6b3e8cf492411a8e5ae3072d44fd5ab349e21951c62b0ed30a9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -180247,9 +180247,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -180259,7 +180259,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -180272,8 +180272,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -180283,11 +180283,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 signed INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping s8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6117422b739d9c39ae85c36a064b6ba0cfe6242cbd6c19fffe7ff68365e00a79.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -180298,15 +180298,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a84718234f047484ec8d86c72e2a333a2982ea25b4266f109070266baa49c715.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -180539,9 +180539,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -180551,7 +180551,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -180564,8 +180564,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -180575,11 +180575,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 signed saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating s8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3c5d9bb6256f7009234918046b2dfb11507265c220722e1a42aa20653a409aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -180590,15 +180590,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9c0a5d19a7e1637ff335da9a214e99e06463d5160df778e1bad3dfcfc6874890.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -180832,9 +180832,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -180844,7 +180844,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -180857,8 +180857,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -180868,11 +180868,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping s8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32f1c01ee1a43ca81812d076ee8db48b8928aa441daf3f71dfc025f6b50a516a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -180883,15 +180883,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 875de648ec423e0f311608024d48bd81ae45ae442c9741a8f08c01502f863e9b.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -181124,9 +181124,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -181136,7 +181136,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -181149,8 +181149,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -181160,11 +181160,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating s8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b79bb4f92abfb0ce7a3b13baed56134e8043e73d4552d305d42f8f2d97661bfd.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -181175,15 +181175,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 adbd36486601138ffe021f493473ae535351c523380e780a11678e1a4da0e4e7.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -181417,9 +181417,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -181429,7 +181429,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -181442,8 +181442,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -181453,11 +181453,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 unsigned INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping u8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34822f024cc535508a1d7ae0c22eb25c61a337e07e674e4c5d0821c85d2e44e0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -181468,15 +181468,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c0cad4768d71256792c04fdf7e962f3630275a62a909d6ecd15a67a1aacb7765.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -181709,9 +181709,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -181721,7 +181721,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -181734,8 +181734,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -181745,11 +181745,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping u8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1dc739b2f8f0bc7f5d2a744a638f9f9e93a9c9763584bdc4ef239e3b740a7900.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -181760,15 +181760,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11793a16a1edc4a91afc40987b925e8829ac10289034931cf23a5dd19c54ff17.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -182001,9 +182001,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -182013,7 +182013,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -182026,8 +182026,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -182037,11 +182037,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating u8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d51f3473017b1bc3eafcae85544debb79cd50a621baebf76491730e6af4dfd83.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -182052,15 +182052,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c8b4cd0226a362e5b29976f9f58f7fba85de7a0f25858de6e77bef814660815e.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -182294,9 +182294,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -182306,7 +182306,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -182319,8 +182319,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -182330,11 +182330,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent sparse m16n8k32 unsigned saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating u8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d8c7f3a382ee9d1ea29abdb323d012394df4bf6bce1fa24e8594c78e8bfa65f7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -182345,15 +182345,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46586ec2c81c1e15a80b74485e9e14cad94c67af27deead4a2ce433787a814af.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -182597,9 +182597,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -182609,7 +182609,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -182622,8 +182622,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -182633,11 +182633,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "four convergent plain-sparsity sparse FP8 F32 inline-PTX probe modules in D, A, B, C, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -182648,14 +182648,14 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "exact-floor plain-sparsity sparse FP8 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4.",
               "artifact_kind": "cubin",
-              "tool_path": "/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -182914,9 +182914,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -182926,7 +182926,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -182939,8 +182939,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -182950,11 +182950,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "four convergent plain-sparsity sparse FP8 F32 inline-PTX probe modules in D, A, B, C, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -182965,14 +182965,14 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "exact-floor plain-sparsity sparse FP8 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4.",
               "artifact_kind": "cubin",
-              "tool_path": "/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -183231,9 +183231,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -183243,7 +183243,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -183256,8 +183256,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -183267,11 +183267,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "four convergent plain-sparsity sparse FP8 F32 inline-PTX probe modules in D, A, B, C, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -183282,14 +183282,14 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "exact-floor plain-sparsity sparse FP8 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4.",
               "artifact_kind": "cubin",
-              "tool_path": "/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -183548,9 +183548,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_89",
@@ -183560,7 +183560,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.4",
             "hardware": {
@@ -183573,8 +183573,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -183584,11 +183584,11 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "four convergent plain-sparsity sparse FP8 F32 inline-PTX probe modules in D, A, B, C, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -183599,14 +183599,14 @@
                 "sm_89",
                 "ptx84"
               ],
-              "representation": "exact-floor plain-sparsity sparse FP8 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4.",
               "artifact_kind": "cubin",
-              "tool_path": "/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
               "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
@@ -183855,9 +183855,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -183867,7 +183867,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -183880,8 +183880,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -183891,11 +183891,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 wrapping s4-by-s4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ae995db003010942ebdf8a541fb71e7a209b14d2035dc9b39ceebb44a20a8b22.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -183906,15 +183906,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d3e0072eed343ab18506217d990ecc273e425433b2c7866bbf1cb84c84c48651.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -184162,9 +184162,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -184174,7 +184174,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -184187,8 +184187,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -184198,11 +184198,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 saturating s4-by-s4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0d1ce7e2d45fa78ac932971fa7ccc0f46e00d454de87b3e1ea926b0b22636b9b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -184213,15 +184213,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f8e6cb0ffef1b7c51d3caa4e5f589992e341d54ef3c6b0f0d1e8ac2581ae5db2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -184470,9 +184470,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -184482,7 +184482,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -184495,8 +184495,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -184506,11 +184506,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 wrapping s4-by-u4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 512623d519f30b4de0fca57417959c5a3e4f6208ac12e3691744f3fbec21cefa.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -184521,15 +184521,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e549ea726768ab22519dacc89c24c3546e6cfcf709f8b53b5a8e86030aa8deb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -184777,9 +184777,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -184789,7 +184789,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -184802,8 +184802,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -184813,11 +184813,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 saturating s4-by-u4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 402335a93decd5671794a0c5ec5e6f09ffd69ed6b26754fac45e7f294b3aac25.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -184828,15 +184828,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11e848685fa3b253ee0b3abb5fee26cf6fb7b3e7f567351483da94bd54ed1afc.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -185093,9 +185093,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -185105,7 +185105,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -185118,8 +185118,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -185129,11 +185129,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 signed INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping s8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcd1b6080f311a524c25afacc87c7db3e52597ff2da06494ecc63a319740ff3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -185144,15 +185144,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7c617c1291bbecb8b6aa72f56558797c56ea20bb7a78fd5769aae5d0f12b720.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -185393,9 +185393,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -185405,7 +185405,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -185418,8 +185418,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -185429,11 +185429,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 signed saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating s8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f1627fbb5b81a5df542173347c1dabb3e3493c579f2d1ec962fbfb8bf88936a7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -185444,15 +185444,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 207e717b6d247cfb221fdbcec7a626708ce5b740bdb21f409d1dbbb439941c5e.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -185694,9 +185694,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -185706,7 +185706,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -185719,8 +185719,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -185730,11 +185730,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping s8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9dcb133309ca12ab7c5fabd0703c014103473203370652bf89f8b5c39de9898.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -185745,15 +185745,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b95b07cec0fa97e908c16d96d9ded6697b9173ef6debe699b291aa890a5d22ee.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -185994,9 +185994,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -186006,7 +186006,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -186019,8 +186019,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -186030,11 +186030,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating s8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 35ffdccb3d2495814cf57898071b7b916a6b6ca5627fa62e179523389496a18d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -186045,15 +186045,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 619c473b81d7936bbd80cd8b1195a5911b756c6209ce85beba9dc3b63c929df0.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -186287,9 +186287,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -186299,7 +186299,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -186312,8 +186312,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -186323,11 +186323,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 wrapping u4-by-u4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 48f8afefe6dcc736f9b1e6750a9aa96bcbe0f1406d48a768f5ec04527b9f8ff2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -186338,15 +186338,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e41e8f8421b3066da06e8b9f77e52d19c248e2833be81e57cbe83b1dc4e3b34a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -186594,9 +186594,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -186606,7 +186606,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -186619,8 +186619,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -186630,11 +186630,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 wrapping u4-by-s4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cfc39a65a680d1cbbc4c17da9f23149ff569b977c427d1b56d9e22ae71752d6d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -186645,15 +186645,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c523935c6e4d65c559d3b3246acf09a9515e0fade43ca29a79225e3009374.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -186901,9 +186901,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -186913,7 +186913,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -186926,8 +186926,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -186937,11 +186937,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 saturating u4-by-s4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c8888ee61f060e895057d138bf3de9ec113eb39bb2c874055eda73abac1316e9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -186952,15 +186952,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 770bae1810b48f7e160fb5a12fc8c49e4075d0c9cf30402bfa707c38346aa36a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -187209,9 +187209,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -187221,7 +187221,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -187234,8 +187234,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -187245,11 +187245,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent standard sparse m16n8k64 saturating u4-by-u4 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d2ab75eca2756a0c4d2c5544fe7837b92630aa42748c757bf68e0dcf6cc0c8f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -187260,15 +187260,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dea3c30a40b3bf4e28518ac19ce412aed32cb5f0394d48bb75ee09953527c0f4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -187525,9 +187525,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -187537,7 +187537,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -187550,8 +187550,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -187561,11 +187561,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 unsigned INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping u8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 73597b682d1b12c2fccfad62f40ece013c3c3fcd9024e59c32ef84a8732364e5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -187576,15 +187576,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ac63519d007e8dce250d4517871606b252765146f7f59abdd1602790cb331543.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -187825,9 +187825,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -187837,7 +187837,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -187850,8 +187850,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -187861,11 +187861,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping u8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c4d93f3a3ddbc1b78370fc6d41ef8514a6a7aabaa7ec338b788f32dd08f910ae.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -187876,15 +187876,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46b4764da009f8939b74a6235f29b9a6fed892e8b86f68ff56d576e2e5d11151.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -188125,9 +188125,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -188137,7 +188137,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -188150,8 +188150,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -188161,11 +188161,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating u8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ee679b12117e5c9162717c0d88a751b96da7576d08049d06c19e3fb6c6300c73.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -188176,15 +188176,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 00174476515d29940f2a3bd1efce00363d2005a85306c962ce53c6a9bc84e43b.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -188426,9 +188426,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -188438,7 +188438,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.1",
             "hardware": {
@@ -188451,8 +188451,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -188462,11 +188462,11 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "convergent base sparse m16n8k64 unsigned saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating u8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a3f0063a7b3cacd62f63384b92d8719e8a4697c12df321ff9a1d5dece0e84d13.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -188477,15 +188477,15 @@
                 "sm_80",
                 "ptx71"
               ],
-              "representation": "exact-floor base sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ff140019dd06530e11bd9507f0e371f07c5d88300c8dbbcd619681437d175825.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -188727,9 +188727,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -188739,7 +188739,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -188752,8 +188752,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -188763,11 +188763,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 5,008-byte combined inline-PTX probe (SHA-256 7145b7febe8ce9f319a106e508a95639d3c06ac7a7cdf1be2c17e0c1f637e516) used public C, A, B, metadata, selector order and covered all eight variants with selector 0. Pinned LLVM emitted 4,025-byte PTX (SHA-256 9eda0f598c87254794f6e3d9e4cae4477b86b5ccb2a0146909c1fed715cce198) with exactly eight ordered m16n8k128 INT4 instructions. An auxiliary 4,800-byte typed LLVM comparison (SHA-256 71c233f005c86cc6616c05c5189171fc62d1525e6e8509927c194177935e8778) used source A, B, C, metadata, selector order and emitted 13,746-byte PTX (SHA-256 5d62e76cc6106809b4a017d82b2b8191dff13107effc9908de13b3a18406eb30) with the same eight forms; this record retained wrapping s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7797f8b2f85799bd268bf06f71383c5e441791fb25d6c11afe368dde7b688f2a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -188778,15 +188778,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight instructions and produced a 4,008-byte cubin with SHA-256 bdab25ef1571d327ffd4bb0023276418b0f44548683778242827997a726ac660 containing exactly eight IMMA.SP.168128 instructions. LLVM and ptxas rejected selector 1 and dynamic selectors; ptxas also rejected PTX 8.4 and sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddf20d448ac238027732e5de25eceff5461797fae6dadac35f3ff3a32a85a89.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -189027,9 +189027,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -189039,7 +189039,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -189052,8 +189052,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -189063,11 +189063,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 468c19abe84ce7bca229d73c81d4ded6564c8d148b95f351a057e04b1b80098e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -189078,15 +189078,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fb8eb364963656a48995d4052deca74cec8dfd07395d74788b843d7264a7128.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -189328,9 +189328,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -189340,7 +189340,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -189353,8 +189353,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -189364,11 +189364,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f1dcbdef6c39dbf4dd3542b151d9ca227d9ba45ff1e57d7f5d5ee7a93f09a0d4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -189379,15 +189379,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cc679e863a7f3653d5ef77a16942b4107959b879360a5b4cf0dd0c3d237cc3f6.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -189628,9 +189628,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -189640,7 +189640,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -189653,8 +189653,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -189664,11 +189664,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5696e1a6c73da853e281aa34b7d6d2fe6dc8be97c211b86c2896a53e6526045a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -189679,15 +189679,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75ac82bc1ef628d35241da2b78df5701adbe94efd1da970a2c40b47deb921515.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -189929,9 +189929,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -189941,7 +189941,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -189954,8 +189954,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -189965,11 +189965,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 943b9548d42ec57747affb74e4de43cd2f8b6dffcc84b3f57c61c014622672e4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -189980,15 +189980,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46e7ba5d6bd8e53adf639c09a6f053a9d08edb608b10649a80c737ced39afaa4.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -190229,9 +190229,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -190241,7 +190241,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -190254,8 +190254,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -190265,11 +190265,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2029ebf4dd778b25d084af4e1383c8952caf0bd2f0c8e5940d2f82be7158e34d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -190280,15 +190280,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4b266bce7d640560951da1ff3eda5e0cc1550c8614b2a1ea6183519c8d722903.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -190529,9 +190529,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -190541,7 +190541,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -190554,8 +190554,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -190565,11 +190565,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b8cf58c826726dd7c44c2c15f2530af1e33cbe393bd65e8de867ef14ed2276c3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -190580,15 +190580,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a19272331578b9876a758d37c2e72dcac32cc7151aa0b2aeb549388f74c8cb51.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -190830,9 +190830,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -190842,7 +190842,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -190855,8 +190855,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -190866,11 +190866,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 193a37484a4f7f25aa08e90c23783e2bbacb3cac45721e59fc60bd86d6b12c79.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -190881,15 +190881,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b9455f8cdc76ec4612f947ce914f722c74032563c11075dad9caf80353c14623.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -191115,10 +191115,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -191127,7 +191127,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -191140,22 +191140,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -191166,15 +191166,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -191184,8 +191184,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -191452,10 +191452,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -191464,7 +191464,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -191477,22 +191477,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -191503,15 +191503,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -191521,8 +191521,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -191789,10 +191789,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -191801,7 +191801,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -191814,22 +191814,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -191840,15 +191840,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -191858,8 +191858,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -192134,10 +192134,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -192146,7 +192146,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -192159,22 +192159,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -192185,15 +192185,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -192203,8 +192203,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -192471,10 +192471,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -192483,7 +192483,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -192496,22 +192496,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -192522,15 +192522,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -192540,8 +192540,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -192816,10 +192816,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -192828,7 +192828,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -192841,22 +192841,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -192867,15 +192867,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -192885,8 +192885,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -193161,10 +193161,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -193173,7 +193173,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -193186,22 +193186,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -193212,15 +193212,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -193230,8 +193230,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -193498,9 +193498,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -193510,7 +193510,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -193523,8 +193523,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -193534,11 +193534,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 signed INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6fd14bda0a191ba64fe0a74811fd81acef22dae9a7364c814feb5dc6b82f25b6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -193549,15 +193549,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33a756465fa9b33d89f2b1ab0da1d5ce7a1f0e8293d97888b8db3e9cdc7a1f4a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -193790,9 +193790,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -193802,7 +193802,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -193815,8 +193815,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -193826,11 +193826,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 signed saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0097b025872046cef28efa3c9f463bf8993bb9c41da22afc8477efc2a17a4ac3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -193841,15 +193841,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75f59d33a1fc63c903fefc73ee06a31dfa1364d3a5dedcaa5cda933152cc92d2.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -194083,9 +194083,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -194095,7 +194095,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -194108,8 +194108,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -194119,11 +194119,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 64edac3534d432072fb0cb9ee94ae758f539d41f98556515b3c8269967feb7a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -194134,15 +194134,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9111691371ac05725c5828e7342d57b2c0dc17a189490d5fe06a4e499c657703.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -194375,9 +194375,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -194387,7 +194387,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -194400,8 +194400,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -194411,11 +194411,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b93a06888a9596b56f9ec7dec0e30eb7a87767aa6c4bab12aac88b0c80e4c7e0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -194426,15 +194426,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6fb8d4d5c0e41f500f6c523bd132db310bb25628a68ae9cb1331f0a890017d87.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -194668,9 +194668,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -194680,7 +194680,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -194693,8 +194693,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -194704,11 +194704,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 unsigned INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cd7e55a90134cfc35729cd87003a44c454b151e8910822841d1d18f0c9ebc1a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -194719,15 +194719,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1566d75ad4047bc212b72c1d4e10c883b8616bb4276963654b1d5a2086c2aa8.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -194960,9 +194960,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -194972,7 +194972,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -194985,8 +194985,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -194996,11 +194996,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12003882bbbb7dab7f4bec48ce3a009d0b9d08e185f69fb277afceef476fb7c9.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195011,15 +195011,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e0029151f55c5f3974fcf5c52523d22289796d7bfbd8fa74916f3cee0a82459d.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -195252,9 +195252,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -195264,7 +195264,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -195277,8 +195277,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -195288,11 +195288,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c10be1e6711b2a58cbe3fa9cf60de15556b9496ba056edee6036a3c20b8dd2d6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195303,15 +195303,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c07f0d6c2eea06bdaa939cf307a9323910485971dd341374537c4067524534c.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -195545,9 +195545,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -195557,7 +195557,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -195570,8 +195570,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -195581,11 +195581,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered-metadata sparse m16n8k32 unsigned saturating INT8 inline PTX with selector immediates 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c30b514d66dd4ef7d3898ed560a1144be98a172ef526ce4388beaf1f7c7dbd44.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195596,15 +195596,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 766d57b5690df60d31209316dc540cfc2f8c31c303b3b3fae6d523385f764f40.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -195849,9 +195849,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -195861,7 +195861,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -195886,8 +195886,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -195897,11 +195897,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195912,11 +195912,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195927,26 +195927,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195957,26 +195957,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -195987,26 +195987,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -196017,15 +196017,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -196432,9 +196432,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -196444,7 +196444,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -196469,8 +196469,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -196480,11 +196480,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -196495,11 +196495,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -196510,26 +196510,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -196540,26 +196540,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -196570,26 +196570,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -196600,15 +196600,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -197015,9 +197015,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -197027,7 +197027,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -197052,8 +197052,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -197063,11 +197063,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197078,11 +197078,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197093,26 +197093,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197123,26 +197123,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197153,26 +197153,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197183,15 +197183,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -197598,9 +197598,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -197610,7 +197610,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -197635,8 +197635,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -197646,11 +197646,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197661,11 +197661,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197676,26 +197676,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197706,26 +197706,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197736,26 +197736,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -197766,15 +197766,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -198181,9 +198181,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -198193,7 +198193,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -198218,8 +198218,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -198229,11 +198229,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198244,11 +198244,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198259,26 +198259,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198289,26 +198289,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198319,26 +198319,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198349,15 +198349,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -198764,9 +198764,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -198776,7 +198776,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -198801,8 +198801,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -198812,11 +198812,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198827,11 +198827,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198842,26 +198842,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198872,26 +198872,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198902,26 +198902,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -198932,15 +198932,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -199347,9 +199347,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -199359,7 +199359,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -199384,8 +199384,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -199395,11 +199395,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -199410,11 +199410,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -199425,26 +199425,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -199455,26 +199455,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -199485,26 +199485,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -199515,15 +199515,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -199930,9 +199930,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -199942,7 +199942,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -199967,8 +199967,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -199978,11 +199978,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -199993,11 +199993,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200008,26 +200008,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200038,26 +200038,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200068,26 +200068,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200098,15 +200098,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -200513,9 +200513,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -200525,7 +200525,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -200550,8 +200550,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -200561,11 +200561,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200576,11 +200576,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200591,26 +200591,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200621,26 +200621,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200651,26 +200651,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -200681,15 +200681,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -201096,9 +201096,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -201108,7 +201108,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -201133,8 +201133,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -201144,11 +201144,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201159,11 +201159,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201174,26 +201174,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201204,26 +201204,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201234,26 +201234,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201264,15 +201264,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -201679,9 +201679,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -201691,7 +201691,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -201716,8 +201716,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -201727,11 +201727,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201742,11 +201742,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201757,26 +201757,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201787,26 +201787,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201817,26 +201817,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -201847,15 +201847,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -202262,9 +202262,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -202274,7 +202274,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -202299,8 +202299,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -202310,11 +202310,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202325,11 +202325,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202340,26 +202340,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202370,26 +202370,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202400,26 +202400,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202430,15 +202430,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -202845,9 +202845,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -202857,7 +202857,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -202882,8 +202882,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -202893,11 +202893,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202908,11 +202908,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202923,26 +202923,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202953,26 +202953,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -202983,26 +202983,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -203013,15 +203013,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -203428,9 +203428,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -203440,7 +203440,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -203465,8 +203465,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -203476,11 +203476,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -203491,11 +203491,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -203506,26 +203506,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -203536,26 +203536,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -203566,26 +203566,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -203596,15 +203596,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -204011,9 +204011,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -204023,7 +204023,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -204048,8 +204048,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -204059,11 +204059,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204074,11 +204074,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204089,26 +204089,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204119,26 +204119,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204149,26 +204149,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204179,15 +204179,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -204594,9 +204594,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -204606,7 +204606,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -204631,8 +204631,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -204642,11 +204642,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204657,11 +204657,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204672,26 +204672,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204702,26 +204702,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204732,26 +204732,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -204762,15 +204762,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -205177,9 +205177,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -205189,7 +205189,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -205214,8 +205214,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -205225,11 +205225,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205240,11 +205240,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205255,26 +205255,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205285,26 +205285,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205315,26 +205315,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205345,15 +205345,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -205760,9 +205760,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -205772,7 +205772,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -205797,8 +205797,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -205808,11 +205808,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205823,11 +205823,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205838,26 +205838,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205868,26 +205868,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205898,26 +205898,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -205928,15 +205928,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -206343,9 +206343,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -206355,7 +206355,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -206380,8 +206380,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -206391,11 +206391,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -206406,11 +206406,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -206421,26 +206421,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -206451,26 +206451,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -206481,26 +206481,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -206511,15 +206511,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -206926,9 +206926,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -206938,7 +206938,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -206963,8 +206963,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -206974,11 +206974,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -206989,11 +206989,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207004,26 +207004,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207034,26 +207034,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207064,26 +207064,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207094,15 +207094,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -207509,9 +207509,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -207521,7 +207521,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -207546,8 +207546,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -207557,11 +207557,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207572,11 +207572,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207587,26 +207587,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207617,26 +207617,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207647,26 +207647,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -207677,15 +207677,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -208092,9 +208092,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -208104,7 +208104,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -208129,8 +208129,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -208140,11 +208140,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208155,11 +208155,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208170,26 +208170,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208200,26 +208200,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208230,26 +208230,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208260,15 +208260,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -208675,9 +208675,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -208687,7 +208687,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -208712,8 +208712,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -208723,11 +208723,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208738,11 +208738,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208753,26 +208753,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208783,26 +208783,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208813,26 +208813,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -208843,15 +208843,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -209258,9 +209258,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -209270,7 +209270,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -209295,8 +209295,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -209306,11 +209306,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209321,11 +209321,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209336,26 +209336,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209366,26 +209366,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209396,26 +209396,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209426,15 +209426,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -209841,9 +209841,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -209853,7 +209853,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -209878,8 +209878,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -209889,11 +209889,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "typed declaration and convergence comparison",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209904,11 +209904,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209919,26 +209919,26 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209949,26 +209949,26 @@
                 "sm_120f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -209979,26 +209979,26 @@
                 "sm_121a",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -210009,15 +210009,15 @@
                 "sm_121f",
                 "ptx88"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -210432,9 +210432,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -210444,7 +210444,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -210469,8 +210469,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -210480,11 +210480,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -210495,15 +210495,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -210783,9 +210783,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -210795,7 +210795,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -210820,8 +210820,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -210831,11 +210831,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -210846,15 +210846,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -211134,9 +211134,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -211146,7 +211146,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -211171,8 +211171,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -211182,11 +211182,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -211197,15 +211197,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -211485,9 +211485,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -211497,7 +211497,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -211522,8 +211522,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -211533,11 +211533,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -211548,15 +211548,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -211836,9 +211836,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -211848,7 +211848,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -211873,8 +211873,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -211884,11 +211884,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -211899,15 +211899,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -212187,9 +212187,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -212199,7 +212199,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -212224,8 +212224,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -212235,11 +212235,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -212250,15 +212250,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -212538,9 +212538,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -212550,7 +212550,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -212575,8 +212575,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -212586,11 +212586,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -212601,15 +212601,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -212889,9 +212889,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -212901,7 +212901,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -212926,8 +212926,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -212937,11 +212937,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -212952,15 +212952,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -213240,9 +213240,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -213252,7 +213252,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -213277,8 +213277,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -213288,11 +213288,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -213303,15 +213303,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -213591,9 +213591,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -213603,7 +213603,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -213628,8 +213628,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -213639,11 +213639,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -213654,15 +213654,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -213942,9 +213942,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -213954,7 +213954,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -213979,8 +213979,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -213990,11 +213990,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -214005,15 +214005,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -214293,9 +214293,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -214305,7 +214305,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -214330,8 +214330,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -214341,11 +214341,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -214356,15 +214356,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -214644,9 +214644,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -214656,7 +214656,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -214681,8 +214681,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -214692,11 +214692,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -214707,15 +214707,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -214995,9 +214995,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -215007,7 +215007,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -215032,8 +215032,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -215043,11 +215043,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -215058,15 +215058,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -215346,9 +215346,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -215358,7 +215358,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -215383,8 +215383,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -215394,11 +215394,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -215409,15 +215409,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -215697,9 +215697,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -215709,7 +215709,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -215734,8 +215734,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -215745,11 +215745,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -215760,15 +215760,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -216048,9 +216048,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -216060,7 +216060,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -216085,8 +216085,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -216096,11 +216096,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -216111,15 +216111,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -216399,9 +216399,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -216411,7 +216411,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -216436,8 +216436,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -216447,11 +216447,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -216462,15 +216462,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -216750,9 +216750,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -216762,7 +216762,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -216787,8 +216787,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -216798,11 +216798,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -216813,15 +216813,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -217101,9 +217101,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -217113,7 +217113,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -217138,8 +217138,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -217149,11 +217149,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -217164,15 +217164,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -217452,9 +217452,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -217464,7 +217464,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -217489,8 +217489,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -217500,11 +217500,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -217515,15 +217515,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -217803,9 +217803,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -217815,7 +217815,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -217840,8 +217840,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -217851,11 +217851,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -217866,15 +217866,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -218154,9 +218154,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -218166,7 +218166,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -218191,8 +218191,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -218202,11 +218202,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -218217,15 +218217,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -218505,9 +218505,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -218517,7 +218517,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -218542,8 +218542,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -218553,11 +218553,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -218568,15 +218568,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -218856,9 +218856,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_120a",
@@ -218868,7 +218868,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.7",
             "hardware": {
@@ -218893,8 +218893,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -218904,11 +218904,11 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -218919,15 +218919,15 @@
                 "sm_120a",
                 "ptx87"
               ],
-              "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+              "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -219188,9 +219188,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -219200,7 +219200,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -219213,8 +219213,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -219224,11 +219224,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 8,107-byte combined inline-PTX probe (SHA-256 ce31bb4c5d7f083ae8e1293a0a3e5196575482427a041d549f9ecd3862e46942) used public C, A, B, metadata, selector order and covered all eight variants with both selector values. Pinned LLVM emitted 8,453-byte PTX (SHA-256 7a8bb0261fd16c1941b47500d70f3cff95a1acb95c51e52480af15a4effc8818) with exactly 16 ordered m16n8k64 INT4 instructions. An auxiliary 6,948-byte typed LLVM comparison (SHA-256 379488c94c4e3d6b1806928d72723845b7bd329857f4dfc3caf8f74d7a5c6649) used source A, B, C, metadata, selector order and emitted 22,330-byte PTX (SHA-256 913b74cc0dba68779655073b83e5a534bcf076fa5e6a1d1d6029ef951b0fc1cd) with the same 16 forms; this record retained wrapping s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34df0473cf3a71f266d58df0459e952a3e93d223f7635e3c0d81fdec210e8e0a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -219239,15 +219239,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 16 instructions and produced a 4,008-byte cubin with SHA-256 a6df23e0492e6f26ede3da7f7fab6c9fdab2693c5c2d6bda12339948cba640ec containing exactly 16 IMMA.SP.16864 instructions. LLVM and ptxas rejected selector 2 and dynamic selectors; ptxas also rejected PTX 8.4 and sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 884148218b4dbe52217fe5aaceae1dbe2a6322f13e2861aba575e5041cdcb472.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -219480,9 +219480,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -219492,7 +219492,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -219505,8 +219505,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -219516,11 +219516,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating s4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ef0b5eecdcc862411ed22bc193ddbcea8d94eb7d7f89158a7764a9759137fbc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -219531,15 +219531,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 782d840202c153255ac2ac63fdee9b5284b1fb992ec0e82e4fb388e11ba09780.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -219773,9 +219773,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -219785,7 +219785,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -219798,8 +219798,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -219809,11 +219809,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e12664cd3c2c8ab1ebee94128e90959dfd9fde36f4023cd59ad4ae6b317a1107.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -219824,15 +219824,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14dcc3ca9825b8567bc0ab8f60df13f8a003993172ec9d8b6463f31eb12b5b1e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -220065,9 +220065,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -220077,7 +220077,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -220090,8 +220090,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -220101,11 +220101,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating s4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 940e92ff27f876255377efa918c901fed7e3e68e5ea77b987c1e267dc681f894.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -220116,15 +220116,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4b51c1e1828a81d1cf6c7fe583ca851ab460871807158be6476b77382dedf9e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -220366,9 +220366,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -220378,7 +220378,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -220391,8 +220391,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -220402,11 +220402,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 signed INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping s8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 978ee747ff8cc5c08d711a0c53853888946906e2669cc24a21928ad8adf71e8e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -220417,15 +220417,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 018e9891184767f6e465cb1f24794c6c0145ef5077c2beeca95e42e05aa4189d.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -220666,9 +220666,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -220678,7 +220678,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -220691,8 +220691,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -220702,11 +220702,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 signed saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating s8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2ae77445ff0a92724257c8a3fb55f6db72266289f6b97e6b47d21516466e5821.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -220717,15 +220717,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e6527d661a654f0d5d7b508d529f9008ea0aa02aca95bc47712d9669b74a4988.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -220967,9 +220967,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -220979,7 +220979,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -220992,8 +220992,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -221003,11 +221003,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping s8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b74c85eca5faf76b32dcda134941e5b09e74cf1aadbfca395a09ad0a4f66286a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -221018,15 +221018,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 041f6695d70e0ae239ca5a008247c1c84c651ecdff45cb9a578559f6c0a27731.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -221267,9 +221267,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -221279,7 +221279,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -221292,8 +221292,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -221303,11 +221303,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating s8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c147cb0046175cd363c824b78ace3b1acdd06bb5c9d49990db97a715a2ef530.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -221318,15 +221318,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3859169c6918f08e624d5ace5c0745797f3757d9f96cad14b37655e71e37e405.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -221560,9 +221560,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -221572,7 +221572,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -221585,8 +221585,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -221596,11 +221596,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1a94dd1d6973c3bba56fbbb5e2972c81d74f9afa874e0e6435de228453839aa4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -221611,15 +221611,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b66dbbbbdb29f1a7d3e5cf07a95446ba1875fee261bb0b3e37206aa5c95c9e4d.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -221852,9 +221852,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -221864,7 +221864,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -221877,8 +221877,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -221888,11 +221888,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 206e31e80c020079ad30648d80a62afaf3564e99b9fc60bce10a6d013fe6c3a2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -221903,15 +221903,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 22fa426e26b6e51a27796d155fc1fe63e70a16bf4b36d73020d02c786769cd5e.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -222144,9 +222144,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -222156,7 +222156,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -222169,8 +222169,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -222180,11 +222180,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating u4-by-s4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f81cee09a2ed0f8c2a26a84a608b4d8363fb89c6da62fbc53ef36e4669d56432.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -222195,15 +222195,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb9ce5c9f7687237a762d8ad90791c121859cc83620ca6ef2755a0ead7e93138.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -222437,9 +222437,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -222449,7 +222449,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -222462,8 +222462,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -222473,11 +222473,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating u4-by-u4 PTX.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06528543fb56e5d6e216226520c60e3fba9c10caee88988d660dc3bf62cfc402.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -222488,15 +222488,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d5cab5865f1616c640a859d6acfd3d704c8a4b6a0504edd360b513973809e3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -222738,9 +222738,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -222750,7 +222750,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -222763,8 +222763,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -222774,11 +222774,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 unsigned INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping u8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4aa071df474c2e447be1941f74662bbb1420c01e4fdab1c071aca9d571863b6f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -222789,15 +222789,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ca07c7e22bb41deec0862f637e40df38b910149068f4abc8a1e480493439e7f3.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -223038,9 +223038,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -223050,7 +223050,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -223063,8 +223063,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -223074,11 +223074,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping u8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 72e7739e891781c3475cf8cacdf67a739e6623660d0f83794818612cd297a82e.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -223089,15 +223089,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a7361b01c024ab1c79cbe8744f497eaeba5eacbe7ead475ba79e21b317abe01f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -223338,9 +223338,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -223350,7 +223350,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -223363,8 +223363,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -223374,11 +223374,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating u8-by-s8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 309f99b05e8c94a7d67043662d247b6d2181ace2fa180f341d891848fe49df87.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -223389,15 +223389,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1677a6dc99df6612f9de3109d52d45b0293847849900d5ed579f651df96031ea.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -223639,9 +223639,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -223651,7 +223651,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -223664,8 +223664,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -223675,11 +223675,11 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "convergent ordered sparse m16n8k64 unsigned saturating INT8 inline PTX with selector 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating u8-by-u8 PTX with selector 0.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b3d041020ef9ee6fb6013cc491495ec4add4a035da239b9d560b1017e0daeda7.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -223690,15 +223690,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2b1813cc3f516d79a19fa139c7fc5d119fc1392db6b1f299fb6ef0ef46df6ca.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -223932,10 +223932,10 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-        "status": "executed",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+        "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
         "ptx_feature": "+ptx85"
@@ -223944,7 +223944,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.5",
             "hardware": {
@@ -223957,22 +223957,22 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
-          "status": "executed",
+          "status": "validated",
           "stages": [
             {
               "targets": [
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -223983,15 +223983,15 @@
                 "sm_80",
                 "ptx85"
               ],
-              "representation": "exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -224001,8 +224001,8 @@
               "representation": "all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle",
               "stage": "runtime",
               "mechanism": "inline_ptx",
-              "outcome": "succeeded",
-              "detail": "Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -224213,9 +224213,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-movmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_75",
@@ -224225,7 +224225,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -224238,8 +224238,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -224249,11 +224249,11 @@
                 "sm_75",
                 "ptx78"
               ],
-              "representation": "one convergent LLVM IR inline-PTX probe for the packed b16 transpose",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned Rust LLVM emitted one exact movmatrix instruction; source SHA-256 b8776ed95d8c8e451deea02d8851a238cfcfe570e286a92c94e5971bf06db963, bitcode SHA-256 caf25ab44ed893d449343707f06af85395e86f41852c32ec1b577be184365aee, and PTX SHA-256 f1e1b7d8ad879abd120d6cb32a5999925af0ed77b3a7144a7dfd8db76ea18246.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4d774d39425c57263a42ec917b671796b75c2e755a81255dfc72a40232e32367.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -224264,15 +224264,15 @@
                 "sm_75",
                 "ptx78"
               ],
-              "representation": "642-byte LLVM-produced PTX module containing movmatrix.sync.aligned.m8n8.trans.b16",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 85c6f2304703e7ee71bd8ff7bbb2dec8730ea51244c37934a0cc838b2d00ab69.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -224419,9 +224419,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -224431,7 +224431,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -224444,8 +224444,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -224455,11 +224455,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "pure inline PTX mul.rn.bf16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact mul.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f4e083050e3dc987284bceea5b0064f86a0c78bca3754d8a92284911bbd7fe78.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -224470,15 +224470,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "PTX emitted by the combined three-form sm_90 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6f92a98210d4280598f8059312562b774c73bae8a2baf9af8252031d8f5ac5b.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -224626,9 +224626,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -224638,7 +224638,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -224651,8 +224651,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -224662,11 +224662,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX mul.rn.f16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0642183f45157615161325bf3fd8c344f4d8efea483a8bf2c111fd9bd7c9dc4f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -224677,15 +224677,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 090bddd2dbb7c5459d46de7c6956de9084a097a5041086b13a2a85322673072d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -224833,9 +224833,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -224845,7 +224845,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -224858,8 +224858,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -224869,11 +224869,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -224884,15 +224884,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -225040,9 +225040,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -225052,7 +225052,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -225065,8 +225065,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -225076,11 +225076,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225091,15 +225091,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -225274,9 +225274,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -225286,7 +225286,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -225299,8 +225299,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -225310,11 +225310,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225325,11 +225325,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225340,15 +225340,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -225539,9 +225539,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -225551,7 +225551,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -225564,8 +225564,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -225575,11 +225575,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225590,11 +225590,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225605,15 +225605,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -225804,9 +225804,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -225816,7 +225816,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -225829,8 +225829,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -225840,11 +225840,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225855,11 +225855,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -225870,15 +225870,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -226070,9 +226070,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -226082,7 +226082,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -226095,8 +226095,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -226106,11 +226106,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226121,11 +226121,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226136,15 +226136,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -226335,9 +226335,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -226347,7 +226347,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -226360,8 +226360,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -226371,11 +226371,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226386,11 +226386,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226401,15 +226401,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -226600,9 +226600,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -226612,7 +226612,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -226625,8 +226625,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -226636,11 +226636,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226651,11 +226651,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226666,15 +226666,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -226866,9 +226866,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -226878,7 +226878,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -226891,8 +226891,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -226902,11 +226902,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226917,11 +226917,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -226932,15 +226932,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -227131,9 +227131,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -227143,7 +227143,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -227156,8 +227156,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -227167,11 +227167,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227182,11 +227182,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227197,15 +227197,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -227396,9 +227396,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -227408,7 +227408,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -227421,8 +227421,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -227432,11 +227432,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227447,11 +227447,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227462,15 +227462,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -227662,9 +227662,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -227674,7 +227674,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -227687,8 +227687,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -227698,11 +227698,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227713,11 +227713,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227728,15 +227728,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -227927,9 +227927,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -227939,7 +227939,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -227952,8 +227952,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -227963,11 +227963,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227978,11 +227978,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -227993,15 +227993,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -228192,9 +228192,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -228204,7 +228204,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -228217,8 +228217,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -228228,11 +228228,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -228243,11 +228243,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -228258,15 +228258,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+              "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -228461,9 +228461,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-nanosleep"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -228473,7 +228473,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.3",
             "hardware": {
@@ -228486,8 +228486,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -228497,11 +228497,11 @@
                 "sm_70",
                 "ptx63"
               ],
-              "representation": "generated nanosleep inline-PTX probe",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM retained exact nanosleep.u32 at the route floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb1f49e5307d0e617ee6cf95da7ec7a582167ac87e14e8bc4d24cb4185ee7f68.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -228512,15 +228512,15 @@
                 "sm_75",
                 "ptx63"
               ],
-              "representation": "the LLVM-produced PTX 6.3 module retargeted to the nearest CUDA 13.3 assembler target",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas no longer accepts sm_70; it accepted the otherwise unchanged PTX 6.3 module for forward-compatible sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 68ce336a8ad5452c31fbff9377bccbcb5540d81a83b904ec1cf2641f607fb34a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -228686,9 +228686,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -228698,7 +228698,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -228711,8 +228711,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -228722,11 +228722,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "pure inline PTX neg.bf16x2 with =r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact neg.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2dca2ee0088dbadc0cb0904d95659a2c0d5e7974da0c5e8aff75f0e290504a1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -228737,15 +228737,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 779e29da03300e6d8024f6c98694b6bbad23c127181f299430b3a6b09c8541b5.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -228887,9 +228887,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -228899,7 +228899,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -228912,8 +228912,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -228923,11 +228923,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX neg.f16x2 with =r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained exact neg.f16x2 inline PTX at PTX 6.0/sm_70. An independent generic v2f16 selection probe emitted neg.f16x2 at sm_53; at sm_52 LLVM emitted scalar sign-bit operations and no packed neg instruction.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e44e1921be6e1f3bc1cc0e885622b5bb8a552876f0f300d6cff83701c2b0a8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -228938,15 +228938,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 emitted by the sm_70 neg.f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 737-byte PTX module at sm_75 and produced a 3,112-byte cubin (SHA-256 0e381d181a5d707df9d2437d636d47bf1f80cdff578b38160e3553bec0bc870a). SASS realizes packed negation as HADD2 with negated half lanes. The PTX 5.0 negative module was rejected because neg.f16/neg.f16x2 requires PTX 6.0 or later.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6abb77861c39cdcaed05cd8512d1b2ca13a32c045cecedb72aa4038fc9c0d6e. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -229107,9 +229107,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nsmid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -229119,7 +229119,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -229132,8 +229132,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -229142,11 +229142,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -229157,11 +229157,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed calls lowering to six exact special-register reads",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -229172,15 +229172,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -229356,9 +229356,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nwarpid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -229368,7 +229368,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -229381,8 +229381,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -229391,11 +229391,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -229406,11 +229406,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed calls lowering to six exact special-register reads",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -229421,15 +229421,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+              "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -229589,9 +229589,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -229601,7 +229601,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -229614,8 +229614,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -229625,11 +229625,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "side-effecting inline PTX atom.global.add.noftz.bf16x2 with =r,l,r,~{memory}",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted the exact native packed bf16x2 global atomic-add instruction.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2537a05bf31f004570ad178a6290d201b70855af40a531e7bb3c4930907b72f.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -229640,15 +229640,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM packed-atomic probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact bf16x2 inline-PTX instruction for sm_90.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5d3e002fed008a798ac7c23848acfa14f680ffa01f2d4c11154a6b127f3b71ce.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -229810,9 +229810,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -229822,7 +229822,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.2",
             "hardware": {
@@ -229835,8 +229835,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -229846,11 +229846,11 @@
                 "sm_70",
                 "ptx62"
               ],
-              "representation": "side-effecting inline PTX atom.global.add.noftz.f16x2 with =r,l,r,~{memory}",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted the exact native packed f16x2 global atomic-add instruction.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7769e680b674aa15719338a21a264d9e6a8bd0e7f4a3f90192b4f5507f2472cd.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -229861,15 +229861,15 @@
                 "sm_75",
                 "ptx62"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM packed-atomic probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas (whose supported floor is sm_75) accepted the sm_70/PTX-6.2 backend output when assembled forward for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a37572911b7e3d97e3785be11b99068c4b530eac6b0a16d6d74d6a0c6a56702. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx62, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -230019,9 +230019,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-pmevent"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -230031,7 +230031,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-debug-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-debug-control",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -230044,8 +230044,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -230055,11 +230055,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one LLVM IR probe with side-effect inline PTX for trap, brkpt, and pmevent 15",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230070,15 +230070,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "1,011-byte LLVM-produced PTX module containing trap, brkpt, and pmevent 15",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -230229,9 +230229,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -230241,7 +230241,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -230254,8 +230254,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -230264,11 +230264,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "11 individually generated typed TMA prefetch declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230279,11 +230279,11 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 typed TMA descriptor and tile-prefetch probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230294,15 +230294,15 @@
                 "sm_90",
                 "ptx80"
               ],
-              "representation": "11 LLVM-produced TMA prefetch PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
+              "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -230464,9 +230464,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -230476,7 +230476,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -230489,8 +230489,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -230499,11 +230499,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230514,11 +230514,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230529,15 +230529,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -230707,9 +230707,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -230719,7 +230719,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -230732,8 +230732,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -230742,11 +230742,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230757,11 +230757,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230772,15 +230772,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -230948,9 +230948,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -230960,7 +230960,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -230973,8 +230973,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -230983,11 +230983,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -230998,11 +230998,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231013,15 +231013,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -231190,9 +231190,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -231202,7 +231202,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -231215,8 +231215,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -231225,11 +231225,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231240,11 +231240,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231255,15 +231255,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -231435,9 +231435,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -231447,7 +231447,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -231460,8 +231460,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -231470,11 +231470,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231485,11 +231485,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231500,15 +231500,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -231676,9 +231676,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -231688,7 +231688,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -231701,8 +231701,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -231711,11 +231711,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231726,11 +231726,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231741,15 +231741,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -231918,9 +231918,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -231930,7 +231930,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -231943,8 +231943,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -231953,11 +231953,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed declarations for all seven PRMT records",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231968,11 +231968,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -231983,15 +231983,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+              "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -232162,9 +232162,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -232174,7 +232174,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -232187,8 +232187,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -232198,11 +232198,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232213,11 +232213,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232228,15 +232228,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -232419,9 +232419,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -232431,7 +232431,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -232444,8 +232444,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -232455,11 +232455,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232470,11 +232470,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232485,15 +232485,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -232676,9 +232676,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -232688,7 +232688,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -232701,8 +232701,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -232712,11 +232712,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232727,11 +232727,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232742,15 +232742,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -232932,9 +232932,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -232944,7 +232944,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -232957,8 +232957,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -232968,11 +232968,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232983,11 +232983,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -232998,15 +232998,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -233188,9 +233188,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -233200,7 +233200,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -233213,8 +233213,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -233224,11 +233224,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -233239,11 +233239,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -233254,15 +233254,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -233445,9 +233445,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -233457,7 +233457,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -233470,8 +233470,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -233481,11 +233481,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -233496,11 +233496,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -233511,15 +233511,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -233701,9 +233701,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -233713,7 +233713,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -233726,8 +233726,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -233737,11 +233737,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -233752,11 +233752,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -233767,15 +233767,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -233957,9 +233957,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -233969,7 +233969,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -233982,8 +233982,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -233993,11 +233993,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234008,11 +234008,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234023,15 +234023,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -234214,9 +234214,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -234226,7 +234226,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -234239,8 +234239,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -234250,11 +234250,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234265,11 +234265,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234280,15 +234280,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -234470,9 +234470,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -234482,7 +234482,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -234495,8 +234495,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -234506,11 +234506,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234521,11 +234521,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234536,15 +234536,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -234726,9 +234726,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -234738,7 +234738,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -234751,8 +234751,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -234762,11 +234762,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234777,11 +234777,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -234792,15 +234792,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -234983,9 +234983,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -234995,7 +234995,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -235008,8 +235008,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -235019,11 +235019,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -235034,11 +235034,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -235049,15 +235049,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -235239,9 +235239,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -235251,7 +235251,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -235264,8 +235264,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -235275,11 +235275,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -235290,11 +235290,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -235305,15 +235305,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -235495,9 +235495,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -235507,7 +235507,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -235520,8 +235520,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -235531,11 +235531,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -235546,11 +235546,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -235561,15 +235561,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -235759,9 +235759,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -235884,9 +235884,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -236020,9 +236020,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -236157,9 +236157,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -236295,9 +236295,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -236420,9 +236420,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -236556,9 +236556,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -236682,9 +236682,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -236818,9 +236818,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -236955,9 +236955,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -237093,9 +237093,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -237218,9 +237218,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -237354,9 +237354,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -237480,9 +237480,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -237605,9 +237605,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -237730,9 +237730,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -237842,9 +237842,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -237854,7 +237854,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -237867,8 +237867,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -237878,11 +237878,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -237893,11 +237893,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -237908,15 +237908,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -238092,9 +238092,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -238104,7 +238104,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -238117,8 +238117,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -238128,11 +238128,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -238143,11 +238143,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -238158,15 +238158,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -238342,9 +238342,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -238354,7 +238354,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -238367,8 +238367,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -238378,11 +238378,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -238393,11 +238393,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -238408,15 +238408,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -238638,9 +238638,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -238650,7 +238650,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -238703,8 +238703,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -238713,11 +238713,11 @@
               "targets": [
                 "sm_90a"
               ],
-              "representation": "typed setmaxnreg decrement declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 a559d3e87b535201a9ae87b367b6bbc3cd6c6fc452574311161fa126b2cabfb7, bitcode SHA-256 6a87d7b67460deeb9fca1720371d2d9f1295620546fc167c64deee03d362543b, and canonical IR SHA-256 b5754b28a81283e193437f849a7e07af5488d484fcb5c3c3d2e9a2d15c193d2d.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be9be58f6dcddd987a8632ca75a2ed6a592f6a65ceff960adc24bcb52bb11db3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -238728,11 +238728,11 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "typed setmaxnreg.dec.sync.aligned.u32 call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted the exact instruction in PTX SHA-256 9758b60ad9ed87b7e0a90f9fd57c3edaa90c7c793621c5f911f0fcaef0a55555.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddfec7d31a47990bd677ada57d94afc3bf77325e79e886219aac161fffc0f896.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -238743,15 +238743,15 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "LLVM-produced setmaxnreg decrement PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 37ad4f1284701ac5101858b0792ed16e833420d4e21cbc1c02896d03019a63a2.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -238989,9 +238989,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-execution-control",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -239001,7 +239001,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-execution-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -239054,8 +239054,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -239064,11 +239064,11 @@
               "targets": [
                 "sm_90a"
               ],
-              "representation": "typed setmaxnreg increment declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 e451448a7385c879e0ca127c25cd038b92bffeacf94d6c3487dbeebbf4f2555e, bitcode SHA-256 00e92b75a1ef024a48fb31297bd9a46d831be64e916d3a522f16bc54d19046a3, and canonical IR SHA-256 7d3888169df2d273686c57ac7b169aaee07b0917f6db7b148a6403130aba3a68.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6648953f506cce13f94612bf6d0da7cd3e075fbc63a6afc01f2ae709fb9a415.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -239079,11 +239079,11 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "typed setmaxnreg.inc.sync.aligned.u32 call",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM emitted the exact instruction in PTX SHA-256 1212a8cf1cbba526030c0f67dbb854c5dd2d24e7cf7977ef2137a14638be0d8a.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4032051215c86d96bb88330e799aec22fb09ad3d232fe483e2346692108ff967.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -239094,15 +239094,15 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "LLVM-produced setmaxnreg increment PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06608466abf4fe979cff6b7e5e3a16a92d0daf5df071239d28e943cd7be29916.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -239368,9 +239368,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -239380,7 +239380,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -239393,8 +239393,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -239404,11 +239404,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.down.f32(i32, f32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fe998b216b523ddba326d894f7fd6c1c811b7b8f6793fd67fa9f32bb5426fe93.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -239419,11 +239419,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.down.f32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.down.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e822e31752229f9d5f8379f58427407f4da67974fdf5cb800973a1d420d99843.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -239434,15 +239434,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.down.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 78a64562571e8dfa2d02c779947163cd4c087bf0fb45cfc9b26b7a9448da1d34. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -239706,9 +239706,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -239718,7 +239718,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -239731,8 +239731,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -239742,11 +239742,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.down.i32(i32, i32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12f77b50bd9a2ac2e167e43862d4925890afbc944ce1be9abf7a42bef834a693.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -239757,11 +239757,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.down.i32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.down.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e868aa2aec22540ec973c73eda5ad85c82e432f562a4956b89fc5ec681ff12c4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -239772,15 +239772,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.down.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 56902b3d564c19bb5e6ffca02b1f0f73c0c6ea1a440d7ba209fe9242e86d6b50. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -239958,9 +239958,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -239970,7 +239970,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -239983,8 +239983,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -239994,11 +239994,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.down.b32 instructions with shared operands, and reassembles i64",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dc004e784b4ad1d4151d0fbbb8c7ab4815437672e35cb37c8408d685938d961.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -240009,15 +240009,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM 64-bit down shuffle probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ac475b6fa0fb52beb3890ebaba00ef224041684fd47bfec82f609768fb0ad71f. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -240268,9 +240268,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -240280,7 +240280,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -240293,8 +240293,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -240304,11 +240304,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.idx.f32(i32, f32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d782063e4b1ae122d58089206151ace8d5b7d4f660a5d5159fb1b770f56b46a0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -240319,11 +240319,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.idx.f32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.idx.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e4775aad2f0dbe6b4f3fc2db7d15985eed28a8adcc2031100d3db5a4c25600a4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -240334,15 +240334,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.idx.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 578376a2dccd0ed45a721f981ae2bf7094d65d90e46479f5569dc8a00354275f. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -240606,9 +240606,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -240618,7 +240618,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -240631,8 +240631,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -240642,11 +240642,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.idx.i32(i32, i32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c8015a293ea78a1394796bab968f8d4b065b433149894cf07283122ab75837d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -240657,11 +240657,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.idx.i32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.idx.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dcc151e0a9e33ce7a059a418221504ddf54bb43d6f1b8927cffd7cdd76c93939.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -240672,15 +240672,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.idx.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0dab85a67df2645eacc4979c404b7be148188a35e648c7e180b82aba4269ae71. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -240858,9 +240858,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -240870,7 +240870,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -240883,8 +240883,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -240894,11 +240894,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.idx.b32 instructions with shared operands, and reassembles i64",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 00eed2a3ce51403c06b4557ca323d691094cf59685823dd73b0b329a2fc40d9a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -240909,15 +240909,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM 64-bit idx shuffle probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1301d100cbe1985f544b748d01b5a92e32173c169be106074f0614f23f7a08d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -241168,9 +241168,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -241180,7 +241180,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -241193,8 +241193,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -241204,11 +241204,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.up.f32(i32, f32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a97a689d300dd7c199c714c01189afb154122960104cfc1a1f1e1180c4e7cf3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -241219,11 +241219,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.up.f32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.up.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a3ed93208637ba99cf7cdbdd4ddbcc7dd13f263fcecc0842c46753c8c841ca34.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -241234,15 +241234,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.up.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 01942dd79db0ae31a8cb526a0bab0e1bbac0a00e9494f7fd2c1ac989af6a68b2. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -241506,9 +241506,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -241518,7 +241518,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -241531,8 +241531,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -241542,11 +241542,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.up.i32(i32, i32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d6fef706af3d2993713934e1f5156c47628a6c943899ad369edf140f7b9f5c57.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -241557,11 +241557,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.up.i32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.up.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e9ac34403d0a172c008ea25d9e62a0786926818eec2efb2091811ddca6c0caf.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -241572,15 +241572,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.up.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 faebc8c202740b059817f861bd3cded8188e653457a4b9791ce87506a1c63314. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -241758,9 +241758,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -241770,7 +241770,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -241783,8 +241783,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -241794,11 +241794,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.up.b32 instructions with shared operands, and reassembles i64",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a83c401208ac85d4c6fe20f942147f8bd8410442b4ce60b86079b138ceac2208.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -241809,15 +241809,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM 64-bit up shuffle probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a85d3dccb30f9491380b1f54913e7261c58667a5c22f195298c6a4f0738f3992. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -242068,9 +242068,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -242080,7 +242080,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -242093,8 +242093,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -242104,11 +242104,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.bfly.f32(i32, f32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa450ce0f0bf68458c30deaa762636db4ffcfbd33d8f8c5d627a057b59740efb.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242119,11 +242119,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.bfly.f32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.bfly.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f67d1796c2477484afa60ecc1be1bc640a045537ba52e1512f1cb17bc46898e4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242134,15 +242134,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.bfly.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4ab01d57f019c3f6994b977e6528bbadebb9db1980ff1e4e57480354c0cb1aa0. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -242406,9 +242406,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -242418,7 +242418,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -242431,8 +242431,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -242442,11 +242442,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cc205d40d08b4db407d73012b4e02bacca951cbf2407578220e1a9993241f7c4.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242457,11 +242457,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.shfl.sync.bfly.i32 calls with register and immediate lane and member-mask operands",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.bfly.b32 operand forms with the fixed clamp.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d1ab9e1df5d932129b785eb98a099581f503fadd7a7e22b4e7621f06dd4ecce.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242472,15 +242472,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing all four reachable shfl.sync.bfly.b32 operand forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f4465879f688e4239d1fc44f8e1c2008923c890cf59e0d1e2cad6653348622d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -242658,9 +242658,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -242670,7 +242670,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -242683,8 +242683,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -242694,11 +242694,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.bfly.b32 instructions with shared operands, and reassembles i64",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a423fb03456e022c274ab0aee6f4480239603e12b7657f78a3cce23264b020e6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242709,15 +242709,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the selected Rust LLVM 64-bit butterfly shuffle probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41a4b0b1e404dde5ac1a6f457e13bde66b3a26ce037c64f3c691fa4666f07d5d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -242895,9 +242895,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sin"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -242907,7 +242907,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -242920,8 +242920,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -242931,11 +242931,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242946,11 +242946,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -242961,15 +242961,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -243145,9 +243145,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sin"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -243157,7 +243157,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -243170,8 +243170,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -243181,11 +243181,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243196,11 +243196,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243211,15 +243211,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -243392,9 +243392,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-smid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -243404,7 +243404,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -243417,8 +243417,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -243428,11 +243428,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "side-effecting inline PTX reads of smid and warpid",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted both volatile location samples in 633-byte PTX with SHA-256 8b6bdea0bf90c68baecd7189ae84ba5c24a0ca32a3a347a1dc72ca8a0a62277f.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ec007869a74a8d4d9bf7be114f276c2c4e26f3e4c34eeaab721362bbf9a1918.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243443,15 +243443,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining smid and warpid reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 2,984-byte cubin with SHA-256 2a1f9febb72d01384c5c1c4583e4511020b2c086144d8b25f6aad354a4c34618.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14f47ee4cc2165c26a56b268fb647e325de01ae94e284257de734b97dea0626a. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -243618,9 +243618,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -243630,7 +243630,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -243643,8 +243643,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -243654,11 +243654,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243669,11 +243669,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243684,15 +243684,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -243874,9 +243874,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -243886,7 +243886,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -243899,8 +243899,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -243910,11 +243910,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243925,11 +243925,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -243940,15 +243940,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -244131,9 +244131,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -244143,7 +244143,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -244156,8 +244156,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -244167,11 +244167,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244182,11 +244182,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244197,15 +244197,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -244387,9 +244387,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -244399,7 +244399,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -244412,8 +244412,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -244423,11 +244423,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244438,11 +244438,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244453,15 +244453,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -244643,9 +244643,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -244655,7 +244655,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -244668,8 +244668,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -244679,11 +244679,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244694,11 +244694,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244709,15 +244709,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -244900,9 +244900,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -244912,7 +244912,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -244925,8 +244925,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -244936,11 +244936,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244951,11 +244951,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -244966,15 +244966,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -245156,9 +245156,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -245168,7 +245168,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -245181,8 +245181,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -245192,11 +245192,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245207,11 +245207,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245222,15 +245222,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -245412,9 +245412,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -245424,7 +245424,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -245437,8 +245437,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -245448,11 +245448,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245463,11 +245463,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245478,15 +245478,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -245669,9 +245669,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -245681,7 +245681,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -245694,8 +245694,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -245705,11 +245705,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245720,11 +245720,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245735,15 +245735,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -245925,9 +245925,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -245937,7 +245937,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -245950,8 +245950,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -245961,11 +245961,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245976,11 +245976,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -245991,15 +245991,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -246181,9 +246181,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -246193,7 +246193,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -246206,8 +246206,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -246217,11 +246217,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -246232,11 +246232,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -246247,15 +246247,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -246438,9 +246438,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -246450,7 +246450,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -246463,8 +246463,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -246474,11 +246474,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -246489,11 +246489,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -246504,15 +246504,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -246694,9 +246694,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -246706,7 +246706,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -246719,8 +246719,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -246730,11 +246730,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -246745,11 +246745,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -246760,15 +246760,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -246950,9 +246950,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -246962,7 +246962,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -246975,8 +246975,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -246986,11 +246986,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247001,11 +247001,11 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247016,15 +247016,15 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "28 typed LLVM routes within the 37-function PTX proof",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+              "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -247208,9 +247208,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -247220,7 +247220,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -247233,8 +247233,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -247243,11 +247243,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "four opaque-pointer typed llvm.nvvm.stmatrix declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247258,11 +247258,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "four typed LLVM probes retaining the x2, x2.trans, x4, and x4.trans stores",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247273,15 +247273,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX 7.8 modules containing the four stmatrix forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -247447,9 +247447,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -247459,7 +247459,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -247472,8 +247472,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -247482,11 +247482,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "four opaque-pointer typed llvm.nvvm.stmatrix declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247497,11 +247497,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "four typed LLVM probes retaining the x2, x2.trans, x4, and x4.trans stores",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247512,15 +247512,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX 7.8 modules containing the four stmatrix forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -247693,9 +247693,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -247705,7 +247705,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -247718,8 +247718,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -247728,11 +247728,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "four opaque-pointer typed llvm.nvvm.stmatrix declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247743,11 +247743,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "four typed LLVM probes retaining the x2, x2.trans, x4, and x4.trans stores",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247758,15 +247758,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX 7.8 modules containing the four stmatrix forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -247938,9 +247938,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -247950,7 +247950,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -247963,8 +247963,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -247973,11 +247973,11 @@
               "targets": [
                 "sm_90"
               ],
-              "representation": "four opaque-pointer typed llvm.nvvm.stmatrix declarations",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -247988,11 +247988,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "four typed LLVM probes retaining the x2, x2.trans, x4, and x4.trans stores",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -248003,15 +248003,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "LLVM-produced PTX 7.8 modules containing the four stmatrix forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+              "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -248156,9 +248156,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90",
@@ -248168,7 +248168,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.8",
             "hardware": {
@@ -248181,8 +248181,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -248192,11 +248192,11 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "pure inline PTX sub.rn.bf16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact sub.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 379f54089c227f714a7a906db098eef11d0ac042db017cbd022236e2f366c440.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -248207,15 +248207,15 @@
                 "sm_90",
                 "ptx78"
               ],
-              "representation": "PTX emitted by the combined three-form sm_90 packed-bf16x2 probe",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e912f52acc5af50229ee7d87a37888ffbfd4dddc236e582d317512c281864d0.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -248363,9 +248363,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -248375,7 +248375,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -248388,8 +248388,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -248399,11 +248399,11 @@
                 "sm_70",
                 "ptx60"
               ],
-              "representation": "pure inline PTX sub.rn.f16x2 with =r,r,r",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 260eb7ac7c16c07d7eb3ac88e498b18d9e9f37b1c763d3dca78075888ea7cce8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -248414,15 +248414,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdc2a8724f570ba9cf569ea9240cf60a937116228c42a5d0d9bdcef1af591d35. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -248570,9 +248570,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -248582,7 +248582,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -248595,8 +248595,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -248606,11 +248606,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -248621,15 +248621,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -248777,9 +248777,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100",
@@ -248789,7 +248789,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -248802,8 +248802,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -248813,11 +248813,11 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -248828,15 +248828,15 @@
                 "sm_100",
                 "ptx86"
               ],
-              "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
+              "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
               "artifact_kind": "cubin",
-              "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-              "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-              "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -249013,9 +249013,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-warp-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -249025,7 +249025,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "6.0",
             "hardware": {
@@ -249038,8 +249038,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -249049,11 +249049,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed llvm.nvvm.bar.warp.sync(i32) declaration",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and preserved its convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d5b275af3bb7cfa86185e4ce735308f68edfb3dd4a71261c8ed80131f8f67b84.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -249064,11 +249064,11 @@
                 "sm_30",
                 "ptx60"
               ],
-              "representation": "typed masked warp barrier calls with register and immediate masks",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact register and immediate bar.warp.sync forms.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e7bf824b7fcf96ef357d4279091d73c4cd95081c9a10994d3a2abc4df06f58d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -249079,15 +249079,15 @@
                 "sm_75",
                 "ptx60"
               ],
-              "representation": "PTX 6.0 containing register and immediate bar.warp.sync forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted both forms and produced a 1,464-byte cubin (SHA-256 df48ead94aa2ad3402456419362350879e99993d4b2ce03a05f695e9d58fbf37).",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 395cf308dc13aa17943c5d073be46dc163c23cd9add8444b68a0dcff491240de. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -249253,9 +249253,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -249265,7 +249265,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -249278,8 +249278,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -249288,11 +249288,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "typed intrinsic llvm.nvvm.barrier.cta.sync.aligned.all(i32)",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "LLVM recognized the declaration and added convergent and nocallback attributes.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 25b383a09f3293d7f5ceef1bc66d8400702d7a5c68d73f24532c371030e69c11.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -249303,11 +249303,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "typed barrier call with fixed ID 0",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The selected Rust LLVM backend emitted exact bar.sync 0 PTX at its oldest supported target profile.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26b139fc6199abb20c9a7d92ddb0a30839fb2bdd3eafe285ec63bb49da4bb3c5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -249318,15 +249318,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "PTX 3.2 containing bar.sync 0",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the emitted PTX when assembled for its supported sm_75 floor.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e287209ed91d6c476b48a3d08399a0911d64f601f3ae07e328db3f975fc2d2e9. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -249460,9 +249460,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-tanh"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_80",
@@ -249472,7 +249472,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "7.0",
             "hardware": {
@@ -249485,8 +249485,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -249496,30 +249496,30 @@
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "3 appended inline-PTX unary math probes (ex2.approx.{f32,ftz.f32}, tanh.approx.f32) attested individually",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_80",
                 "ptx70"
               ],
-              "representation": "3 appended inline-PTX scalar math instructions assembled at the admission floor",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -249716,9 +249716,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -249728,7 +249728,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -249753,8 +249753,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -249764,11 +249764,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -249779,15 +249779,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -249980,9 +249980,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -249992,7 +249992,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -250017,8 +250017,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -250028,11 +250028,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -250043,15 +250043,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -250232,9 +250232,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -250244,7 +250244,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -250269,8 +250269,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -250280,11 +250280,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -250295,15 +250295,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -250479,9 +250479,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -250491,7 +250491,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -250516,8 +250516,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -250527,11 +250527,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -250542,15 +250542,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -250737,9 +250737,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -250749,7 +250749,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -250774,8 +250774,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -250785,15 +250785,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "three convergent inline-PTX kernels covering the admitted tcgen05 commit and shift controls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -250803,8 +250803,8 @@
               "representation": "isolated direct-intrinsic probes at and outside the imported selection predicates",
               "stage": "backend_codegen",
               "mechanism": null,
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -250815,15 +250815,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing the three exact tcgen05 controls",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -250833,8 +250833,8 @@
               "representation": "the same three inline instructions emitted for excluded family and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -251075,9 +251075,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -251087,7 +251087,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -251112,8 +251112,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -251123,11 +251123,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -251138,15 +251138,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -251335,9 +251335,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -251347,7 +251347,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -251372,8 +251372,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -251383,11 +251383,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -251398,15 +251398,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -251591,9 +251591,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -251603,7 +251603,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -251628,8 +251628,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -251639,11 +251639,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -251654,15 +251654,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -251850,9 +251850,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -251862,7 +251862,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -251887,8 +251887,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -251898,11 +251898,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -251913,15 +251913,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -252129,9 +252129,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -252141,7 +252141,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -252166,8 +252166,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -252177,11 +252177,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -252192,15 +252192,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -252410,9 +252410,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -252422,7 +252422,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -252447,8 +252447,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -252458,11 +252458,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -252473,15 +252473,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -252691,9 +252691,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -252703,7 +252703,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -252728,8 +252728,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -252739,11 +252739,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -252754,15 +252754,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -252972,9 +252972,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -252984,7 +252984,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -253009,8 +253009,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -253020,11 +253020,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -253035,15 +253035,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -253253,9 +253253,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -253265,7 +253265,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -253290,8 +253290,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -253301,11 +253301,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -253316,15 +253316,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -253532,9 +253532,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -253544,7 +253544,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -253569,8 +253569,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -253580,11 +253580,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -253595,15 +253595,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -253813,9 +253813,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -253825,7 +253825,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -253850,8 +253850,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -253861,11 +253861,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -253876,15 +253876,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -254094,9 +254094,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -254106,7 +254106,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -254131,8 +254131,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -254142,11 +254142,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -254157,15 +254157,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -254375,9 +254375,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -254387,7 +254387,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -254412,8 +254412,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -254423,11 +254423,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -254438,15 +254438,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -254656,9 +254656,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -254668,7 +254668,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -254693,8 +254693,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -254704,11 +254704,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -254719,15 +254719,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -254936,9 +254936,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -254948,7 +254948,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -254973,8 +254973,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -254984,11 +254984,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -254999,15 +254999,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -255218,9 +255218,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -255230,7 +255230,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -255255,8 +255255,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -255266,11 +255266,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -255281,15 +255281,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -255500,9 +255500,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -255512,7 +255512,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -255537,8 +255537,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -255548,11 +255548,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -255563,15 +255563,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -255782,9 +255782,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -255794,7 +255794,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -255819,8 +255819,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -255830,11 +255830,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -255845,15 +255845,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -256064,9 +256064,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -256076,7 +256076,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -256101,8 +256101,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -256112,11 +256112,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -256127,15 +256127,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -256344,9 +256344,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -256356,7 +256356,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -256381,8 +256381,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -256392,11 +256392,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -256407,15 +256407,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -256623,9 +256623,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -256635,7 +256635,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -256660,8 +256660,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -256671,11 +256671,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -256686,15 +256686,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -256904,9 +256904,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -256916,7 +256916,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -256941,8 +256941,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -256952,11 +256952,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -256967,15 +256967,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -257185,9 +257185,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -257197,7 +257197,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -257222,8 +257222,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -257233,11 +257233,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -257248,15 +257248,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -257466,9 +257466,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -257478,7 +257478,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -257503,8 +257503,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -257514,11 +257514,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -257529,15 +257529,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -257747,9 +257747,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -257759,7 +257759,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -257784,8 +257784,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -257795,11 +257795,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -257810,15 +257810,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -258026,9 +258026,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -258038,7 +258038,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -258063,8 +258063,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -258074,11 +258074,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -258089,15 +258089,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -258306,9 +258306,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -258318,7 +258318,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -258343,8 +258343,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -258354,11 +258354,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -258369,15 +258369,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -258588,9 +258588,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -258600,7 +258600,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -258625,8 +258625,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -258636,11 +258636,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -258651,15 +258651,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -258870,9 +258870,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -258882,7 +258882,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -258907,8 +258907,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -258918,11 +258918,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -258933,15 +258933,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -259152,9 +259152,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -259164,7 +259164,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -259189,8 +259189,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -259200,11 +259200,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -259215,15 +259215,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -259434,9 +259434,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -259446,7 +259446,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -259471,8 +259471,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -259482,11 +259482,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -259497,15 +259497,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -259714,9 +259714,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -259726,7 +259726,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -259751,8 +259751,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -259762,11 +259762,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -259777,15 +259777,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -259994,9 +259994,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -260006,7 +260006,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -260031,8 +260031,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -260042,11 +260042,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -260057,15 +260057,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -260276,9 +260276,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -260288,7 +260288,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -260313,8 +260313,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -260324,11 +260324,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -260339,15 +260339,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -260558,9 +260558,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -260570,7 +260570,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -260595,8 +260595,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -260606,11 +260606,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -260621,15 +260621,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -260840,9 +260840,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -260852,7 +260852,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -260877,8 +260877,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -260888,11 +260888,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -260903,15 +260903,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -261122,9 +261122,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -261134,7 +261134,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -261159,8 +261159,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -261170,11 +261170,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -261185,15 +261185,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+              "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -261402,9 +261402,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -261414,7 +261414,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -261439,8 +261439,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -261450,11 +261450,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -261465,15 +261465,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -261662,9 +261662,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -261674,7 +261674,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -261699,8 +261699,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -261710,11 +261710,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -261725,15 +261725,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -261922,9 +261922,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -261934,7 +261934,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -261959,8 +261959,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -261970,11 +261970,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -261985,15 +261985,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -262184,9 +262184,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -262196,7 +262196,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -262221,8 +262221,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -262232,11 +262232,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -262247,15 +262247,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -262436,9 +262436,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -262448,7 +262448,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -262473,8 +262473,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -262484,11 +262484,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -262499,15 +262499,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -262677,9 +262677,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -262689,7 +262689,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -262714,8 +262714,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -262725,11 +262725,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -262740,15 +262740,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -262954,9 +262954,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -262966,7 +262966,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -262991,8 +262991,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -263002,11 +263002,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263020,8 +263020,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263032,15 +263032,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -263050,8 +263050,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -263295,9 +263295,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -263307,7 +263307,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -263332,8 +263332,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -263343,11 +263343,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263361,8 +263361,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263373,15 +263373,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -263391,8 +263391,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -263605,9 +263605,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -263617,7 +263617,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -263642,8 +263642,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -263653,11 +263653,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263671,8 +263671,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263683,15 +263683,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -263701,8 +263701,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -263916,9 +263916,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -263928,7 +263928,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -263953,8 +263953,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -263964,11 +263964,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263982,8 +263982,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -263994,15 +263994,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -264012,8 +264012,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -264228,9 +264228,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -264240,7 +264240,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -264265,8 +264265,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -264276,11 +264276,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -264294,8 +264294,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -264306,15 +264306,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -264324,8 +264324,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -264541,9 +264541,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -264553,7 +264553,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -264578,8 +264578,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -264589,11 +264589,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -264607,8 +264607,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -264619,15 +264619,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -264637,8 +264637,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -264913,9 +264913,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -264925,7 +264925,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -264950,8 +264950,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -264961,11 +264961,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -264979,8 +264979,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -264991,15 +264991,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -265009,8 +265009,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -265286,9 +265286,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -265298,7 +265298,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -265323,8 +265323,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -265334,11 +265334,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -265352,8 +265352,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -265364,15 +265364,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -265382,8 +265382,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -265602,9 +265602,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -265614,7 +265614,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -265639,8 +265639,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -265650,11 +265650,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -265668,8 +265668,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -265680,15 +265680,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -265698,8 +265698,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -265919,9 +265919,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -265931,7 +265931,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -265956,8 +265956,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -265967,11 +265967,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -265985,8 +265985,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -265997,15 +265997,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -266015,8 +266015,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -266355,9 +266355,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -266367,7 +266367,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -266392,8 +266392,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -266403,11 +266403,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -266421,8 +266421,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -266433,15 +266433,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -266451,8 +266451,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -266792,9 +266792,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -266804,7 +266804,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -266829,8 +266829,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -266840,11 +266840,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -266858,8 +266858,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -266870,15 +266870,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -266888,8 +266888,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -267116,9 +267116,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -267128,7 +267128,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -267153,8 +267153,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -267164,11 +267164,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -267182,8 +267182,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -267194,15 +267194,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -267212,8 +267212,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -267441,9 +267441,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -267453,7 +267453,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -267478,8 +267478,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -267489,11 +267489,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -267507,8 +267507,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -267519,15 +267519,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -267537,8 +267537,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -267753,9 +267753,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -267765,7 +267765,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -267790,8 +267790,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -267801,11 +267801,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -267816,15 +267816,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -268075,9 +268075,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -268087,7 +268087,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -268112,8 +268112,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -268123,11 +268123,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -268141,8 +268141,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -268153,15 +268153,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -268171,8 +268171,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -268448,9 +268448,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -268460,7 +268460,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -268485,8 +268485,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -268496,11 +268496,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -268514,8 +268514,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -268526,15 +268526,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -268544,8 +268544,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -268760,9 +268760,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -268772,7 +268772,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -268797,8 +268797,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -268808,11 +268808,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -268826,8 +268826,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -268838,15 +268838,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -268856,8 +268856,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -269073,9 +269073,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -269085,7 +269085,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -269110,8 +269110,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -269121,11 +269121,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -269139,8 +269139,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -269151,15 +269151,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -269169,8 +269169,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -269389,9 +269389,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -269401,7 +269401,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -269426,8 +269426,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -269437,11 +269437,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -269455,8 +269455,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -269467,15 +269467,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -269485,8 +269485,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -269706,9 +269706,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -269718,7 +269718,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -269743,8 +269743,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -269754,11 +269754,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -269772,8 +269772,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -269784,15 +269784,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -269802,8 +269802,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -270142,9 +270142,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -270154,7 +270154,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -270179,8 +270179,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -270190,11 +270190,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -270208,8 +270208,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -270220,15 +270220,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -270238,8 +270238,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -270579,9 +270579,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -270591,7 +270591,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -270616,8 +270616,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -270627,11 +270627,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -270645,8 +270645,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -270657,15 +270657,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -270675,8 +270675,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -270903,9 +270903,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -270915,7 +270915,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -270940,8 +270940,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -270951,11 +270951,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -270969,8 +270969,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -270981,15 +270981,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -270999,8 +270999,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -271228,9 +271228,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -271240,7 +271240,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -271265,8 +271265,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -271276,11 +271276,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -271294,8 +271294,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -271306,15 +271306,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -271324,8 +271324,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -271568,9 +271568,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -271580,7 +271580,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -271605,8 +271605,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -271616,11 +271616,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -271634,8 +271634,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -271646,15 +271646,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -271664,8 +271664,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -271909,9 +271909,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -271921,7 +271921,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -271946,8 +271946,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -271957,11 +271957,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -271972,15 +271972,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -272199,9 +272199,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -272211,7 +272211,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -272236,8 +272236,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -272247,11 +272247,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -272265,8 +272265,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -272277,15 +272277,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -272295,8 +272295,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -272639,9 +272639,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -272651,7 +272651,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -272676,8 +272676,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -272687,15 +272687,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -272705,8 +272705,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -272717,15 +272717,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -272735,8 +272735,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -273098,9 +273098,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -273110,7 +273110,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -273135,8 +273135,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -273146,15 +273146,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -273164,8 +273164,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -273176,15 +273176,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -273194,8 +273194,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -273444,9 +273444,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -273456,7 +273456,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -273481,8 +273481,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -273492,15 +273492,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -273510,8 +273510,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -273522,15 +273522,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -273540,8 +273540,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -273791,9 +273791,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -273803,7 +273803,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -273828,8 +273828,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -273839,15 +273839,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -273857,8 +273857,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -273869,15 +273869,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -273887,8 +273887,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -274122,9 +274122,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -274134,7 +274134,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -274159,8 +274159,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -274170,15 +274170,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -274188,8 +274188,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -274200,15 +274200,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -274218,8 +274218,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -274454,9 +274454,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -274466,7 +274466,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -274491,8 +274491,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -274502,15 +274502,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -274520,8 +274520,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -274532,15 +274532,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -274550,8 +274550,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -274786,9 +274786,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -274798,7 +274798,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -274823,8 +274823,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -274834,15 +274834,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -274852,8 +274852,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -274864,15 +274864,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -274882,8 +274882,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -275119,9 +275119,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -275131,7 +275131,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -275156,8 +275156,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -275167,15 +275167,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -275185,8 +275185,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -275197,15 +275197,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -275215,8 +275215,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -275481,9 +275481,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -275493,7 +275493,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -275518,8 +275518,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -275529,15 +275529,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -275547,8 +275547,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -275559,15 +275559,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -275577,8 +275577,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -275844,9 +275844,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -275856,7 +275856,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -275881,8 +275881,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -275892,15 +275892,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -275910,8 +275910,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -275922,15 +275922,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -275940,8 +275940,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -276178,9 +276178,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -276190,7 +276190,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -276215,8 +276215,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -276226,15 +276226,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -276244,8 +276244,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -276256,15 +276256,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -276274,8 +276274,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -276513,9 +276513,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -276525,7 +276525,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -276550,8 +276550,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -276561,15 +276561,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -276579,8 +276579,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -276591,15 +276591,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -276609,8 +276609,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -276907,9 +276907,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -276919,7 +276919,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -276944,8 +276944,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -276955,15 +276955,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -276973,8 +276973,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -276985,15 +276985,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -277003,8 +277003,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -277302,9 +277302,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -277314,7 +277314,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -277339,8 +277339,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -277350,15 +277350,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -277368,8 +277368,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -277380,15 +277380,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -277398,8 +277398,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -277640,9 +277640,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -277652,7 +277652,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -277677,8 +277677,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -277688,15 +277688,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -277706,8 +277706,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -277718,15 +277718,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -277736,8 +277736,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -277979,9 +277979,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -277991,7 +277991,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -278016,8 +278016,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -278027,15 +278027,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -278045,8 +278045,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -278057,15 +278057,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -278075,8 +278075,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -278433,9 +278433,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -278445,7 +278445,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -278470,8 +278470,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -278481,11 +278481,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -278499,8 +278499,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -278511,15 +278511,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -278529,8 +278529,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -278870,9 +278870,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -278882,7 +278882,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -278907,8 +278907,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -278918,11 +278918,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -278936,8 +278936,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -278948,15 +278948,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -278966,8 +278966,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -279194,9 +279194,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -279206,7 +279206,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -279231,8 +279231,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -279242,11 +279242,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -279260,8 +279260,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -279272,15 +279272,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -279290,8 +279290,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -279519,9 +279519,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -279531,7 +279531,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -279556,8 +279556,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -279567,11 +279567,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -279585,8 +279585,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -279597,15 +279597,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -279615,8 +279615,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -279828,9 +279828,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -279840,7 +279840,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -279865,8 +279865,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -279876,11 +279876,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -279894,8 +279894,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -279906,15 +279906,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -279924,8 +279924,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -280138,9 +280138,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -280150,7 +280150,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -280175,8 +280175,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -280186,11 +280186,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -280204,8 +280204,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -280216,15 +280216,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -280234,8 +280234,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -280448,9 +280448,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -280460,7 +280460,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -280485,8 +280485,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -280496,11 +280496,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -280514,8 +280514,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -280526,15 +280526,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -280544,8 +280544,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -280759,9 +280759,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -280771,7 +280771,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -280796,8 +280796,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -280807,11 +280807,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -280825,8 +280825,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -280837,15 +280837,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -280855,8 +280855,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -281099,9 +281099,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -281111,7 +281111,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -281136,8 +281136,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -281147,11 +281147,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -281165,8 +281165,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -281177,15 +281177,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -281195,8 +281195,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -281440,9 +281440,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -281452,7 +281452,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -281477,8 +281477,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -281488,11 +281488,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -281506,8 +281506,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -281518,15 +281518,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -281536,8 +281536,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -281752,9 +281752,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -281764,7 +281764,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -281789,8 +281789,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -281800,11 +281800,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -281818,8 +281818,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -281830,15 +281830,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -281848,8 +281848,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -282065,9 +282065,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -282077,7 +282077,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -282102,8 +282102,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -282113,11 +282113,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -282131,8 +282131,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -282143,15 +282143,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -282161,8 +282161,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -282437,9 +282437,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -282449,7 +282449,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -282474,8 +282474,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -282485,11 +282485,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -282503,8 +282503,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -282515,15 +282515,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -282533,8 +282533,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -282810,9 +282810,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -282822,7 +282822,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -282847,8 +282847,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -282858,11 +282858,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -282876,8 +282876,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -282888,15 +282888,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -282906,8 +282906,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -283126,9 +283126,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -283138,7 +283138,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -283163,8 +283163,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -283174,11 +283174,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -283192,8 +283192,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -283204,15 +283204,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -283222,8 +283222,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -283443,9 +283443,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -283455,7 +283455,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -283480,8 +283480,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -283491,11 +283491,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -283509,8 +283509,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -283521,15 +283521,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -283539,8 +283539,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -283879,9 +283879,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -283891,7 +283891,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -283916,8 +283916,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -283927,11 +283927,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -283945,8 +283945,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -283957,15 +283957,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -283975,8 +283975,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -284316,9 +284316,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -284328,7 +284328,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -284353,8 +284353,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -284364,11 +284364,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -284382,8 +284382,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -284394,15 +284394,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -284412,8 +284412,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -284640,9 +284640,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -284652,7 +284652,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -284677,8 +284677,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -284688,11 +284688,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -284706,8 +284706,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -284718,15 +284718,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -284736,8 +284736,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -284965,9 +284965,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -284977,7 +284977,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -285002,8 +285002,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -285013,11 +285013,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285031,8 +285031,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285043,15 +285043,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -285061,8 +285061,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -285274,9 +285274,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -285286,7 +285286,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -285311,8 +285311,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -285322,11 +285322,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285340,8 +285340,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285352,15 +285352,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -285370,8 +285370,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -285584,9 +285584,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -285596,7 +285596,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -285621,8 +285621,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -285632,11 +285632,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285650,8 +285650,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285662,15 +285662,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -285680,8 +285680,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -285894,9 +285894,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -285906,7 +285906,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -285931,8 +285931,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -285942,11 +285942,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285960,8 +285960,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -285972,15 +285972,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -285990,8 +285990,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -286205,9 +286205,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -286217,7 +286217,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -286242,8 +286242,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -286253,11 +286253,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -286271,8 +286271,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -286283,15 +286283,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -286301,8 +286301,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -286545,9 +286545,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -286557,7 +286557,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -286582,8 +286582,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -286593,11 +286593,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -286611,8 +286611,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -286623,15 +286623,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -286641,8 +286641,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -286886,9 +286886,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -286898,7 +286898,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -286923,8 +286923,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -286934,11 +286934,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -286952,8 +286952,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -286964,15 +286964,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -286982,8 +286982,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -287198,9 +287198,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -287210,7 +287210,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -287235,8 +287235,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -287246,11 +287246,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -287264,8 +287264,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -287276,15 +287276,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -287294,8 +287294,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -287511,9 +287511,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -287523,7 +287523,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -287548,8 +287548,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -287559,11 +287559,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -287577,8 +287577,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -287589,15 +287589,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -287607,8 +287607,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -287883,9 +287883,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -287895,7 +287895,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -287920,8 +287920,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -287931,11 +287931,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -287949,8 +287949,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -287961,15 +287961,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -287979,8 +287979,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -288256,9 +288256,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -288268,7 +288268,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -288293,8 +288293,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -288304,11 +288304,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -288322,8 +288322,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -288334,15 +288334,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -288352,8 +288352,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -288572,9 +288572,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -288584,7 +288584,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -288609,8 +288609,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -288620,11 +288620,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -288638,8 +288638,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -288650,15 +288650,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -288668,8 +288668,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -288889,9 +288889,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -288901,7 +288901,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -288926,8 +288926,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -288937,11 +288937,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -288955,8 +288955,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -288967,15 +288967,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -288985,8 +288985,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -289193,9 +289193,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -289205,7 +289205,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -289230,8 +289230,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -289241,11 +289241,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289256,15 +289256,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -289513,9 +289513,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -289525,7 +289525,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -289600,8 +289600,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -289611,26 +289611,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289641,11 +289641,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289656,11 +289656,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289671,11 +289671,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289686,11 +289686,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289701,11 +289701,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289716,11 +289716,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289734,8 +289734,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289749,8 +289749,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289764,8 +289764,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -289776,90 +289776,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -289869,8 +289869,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -290616,9 +290616,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -290628,7 +290628,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -290703,8 +290703,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -290714,26 +290714,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290744,11 +290744,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290759,11 +290759,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290774,11 +290774,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290789,11 +290789,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290804,11 +290804,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290819,11 +290819,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290837,8 +290837,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290852,8 +290852,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290867,8 +290867,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -290879,90 +290879,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -290972,8 +290972,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -291719,9 +291719,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -291731,7 +291731,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -291806,8 +291806,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -291817,26 +291817,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291847,11 +291847,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291862,11 +291862,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291877,11 +291877,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291892,11 +291892,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291907,11 +291907,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291922,11 +291922,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291940,8 +291940,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291955,8 +291955,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291970,8 +291970,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -291982,90 +291982,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -292075,8 +292075,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -292822,9 +292822,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -292834,7 +292834,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -292909,8 +292909,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -292920,26 +292920,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -292950,11 +292950,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -292965,11 +292965,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -292980,11 +292980,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -292995,11 +292995,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -293010,11 +293010,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -293025,11 +293025,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -293043,8 +293043,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -293058,8 +293058,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -293073,8 +293073,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -293085,90 +293085,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -293178,8 +293178,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -293925,9 +293925,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -293937,7 +293937,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -294012,8 +294012,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -294023,26 +294023,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294053,11 +294053,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294068,11 +294068,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294083,11 +294083,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294098,11 +294098,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294113,11 +294113,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294128,11 +294128,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294146,8 +294146,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294161,8 +294161,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294176,8 +294176,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -294188,90 +294188,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -294281,8 +294281,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -294968,9 +294968,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -294980,7 +294980,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -295005,8 +295005,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -295016,11 +295016,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -295031,15 +295031,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -295249,9 +295249,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -295261,7 +295261,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -295286,8 +295286,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -295297,11 +295297,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -295312,15 +295312,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -295973,9 +295973,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -295985,7 +295985,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -296223,8 +296223,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -296234,26 +296234,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296264,11 +296264,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296279,11 +296279,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296294,11 +296294,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296309,11 +296309,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296324,11 +296324,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296339,11 +296339,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296357,8 +296357,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296372,8 +296372,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296387,8 +296387,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -296399,90 +296399,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -296492,8 +296492,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -298038,9 +298038,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -298050,7 +298050,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -298288,8 +298288,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -298299,26 +298299,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298329,11 +298329,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298344,11 +298344,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298359,11 +298359,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298374,11 +298374,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298389,11 +298389,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298404,11 +298404,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298422,8 +298422,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298437,8 +298437,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298452,8 +298452,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -298464,90 +298464,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -298557,8 +298557,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -300108,9 +300108,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -300120,7 +300120,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -300358,8 +300358,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -300369,26 +300369,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300399,11 +300399,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300414,11 +300414,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300429,11 +300429,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300444,11 +300444,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300459,11 +300459,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300474,11 +300474,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300492,8 +300492,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300507,8 +300507,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300522,8 +300522,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -300534,90 +300534,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -300627,8 +300627,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -302066,9 +302066,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -302078,7 +302078,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -302316,8 +302316,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -302327,26 +302327,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302357,11 +302357,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302372,11 +302372,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302387,11 +302387,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302402,11 +302402,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302417,11 +302417,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302432,11 +302432,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302450,8 +302450,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302465,8 +302465,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302480,8 +302480,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -302492,90 +302492,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -302585,8 +302585,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -304137,9 +304137,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -304149,7 +304149,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -304387,8 +304387,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -304398,26 +304398,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304428,11 +304428,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304443,11 +304443,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304458,11 +304458,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304473,11 +304473,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304488,11 +304488,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304503,11 +304503,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304521,8 +304521,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304536,8 +304536,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304551,8 +304551,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -304563,90 +304563,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -304656,8 +304656,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -306088,9 +306088,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -306100,7 +306100,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -306338,8 +306338,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -306349,26 +306349,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306379,11 +306379,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306394,11 +306394,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306409,11 +306409,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306424,11 +306424,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306439,11 +306439,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306454,11 +306454,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306472,8 +306472,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306487,8 +306487,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306502,8 +306502,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -306514,90 +306514,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -306607,8 +306607,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -307713,9 +307713,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -307725,7 +307725,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -307750,8 +307750,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -307761,11 +307761,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -307776,15 +307776,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -308053,9 +308053,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -308065,7 +308065,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -308140,8 +308140,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -308151,26 +308151,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308181,11 +308181,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308196,11 +308196,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308211,11 +308211,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308226,11 +308226,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308241,11 +308241,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308256,11 +308256,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308274,8 +308274,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308289,8 +308289,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308304,8 +308304,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -308316,90 +308316,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -308409,8 +308409,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -309158,9 +309158,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -309170,7 +309170,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -309245,8 +309245,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -309256,26 +309256,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309286,11 +309286,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309301,11 +309301,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309316,11 +309316,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309331,11 +309331,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309346,11 +309346,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309361,11 +309361,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309379,8 +309379,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309394,8 +309394,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309409,8 +309409,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -309421,90 +309421,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -309514,8 +309514,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -310263,9 +310263,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -310275,7 +310275,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -310350,8 +310350,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -310361,26 +310361,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310391,11 +310391,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310406,11 +310406,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310421,11 +310421,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310436,11 +310436,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310451,11 +310451,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310466,11 +310466,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310484,8 +310484,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310499,8 +310499,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310514,8 +310514,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -310526,90 +310526,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -310619,8 +310619,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -311368,9 +311368,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -311380,7 +311380,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -311455,8 +311455,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -311466,26 +311466,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311496,11 +311496,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311511,11 +311511,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311526,11 +311526,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311541,11 +311541,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311556,11 +311556,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311571,11 +311571,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311589,8 +311589,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311604,8 +311604,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311619,8 +311619,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -311631,90 +311631,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -311724,8 +311724,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -312473,9 +312473,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -312485,7 +312485,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -312560,8 +312560,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -312571,26 +312571,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312601,11 +312601,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312616,11 +312616,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312631,11 +312631,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312646,11 +312646,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312661,11 +312661,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312676,11 +312676,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312694,8 +312694,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312709,8 +312709,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312724,8 +312724,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -312736,90 +312736,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -312829,8 +312829,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -313521,9 +313521,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -313533,7 +313533,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -313558,8 +313558,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -313569,11 +313569,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -313584,15 +313584,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -314466,9 +314466,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -314478,7 +314478,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -314716,8 +314716,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -314727,26 +314727,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314757,11 +314757,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314772,11 +314772,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314787,11 +314787,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314802,11 +314802,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314817,11 +314817,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314832,11 +314832,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314850,8 +314850,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314865,8 +314865,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314880,8 +314880,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -314892,90 +314892,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -314985,8 +314985,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -316755,9 +316755,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -316767,7 +316767,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -317005,8 +317005,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -317016,26 +317016,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317046,11 +317046,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317061,11 +317061,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317076,11 +317076,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317091,11 +317091,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317106,11 +317106,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317121,11 +317121,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317139,8 +317139,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317154,8 +317154,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317169,8 +317169,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -317181,90 +317181,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -317274,8 +317274,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -319047,9 +319047,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -319059,7 +319059,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -319297,8 +319297,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -319308,26 +319308,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319338,11 +319338,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319353,11 +319353,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319368,11 +319368,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319383,11 +319383,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319398,11 +319398,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319413,11 +319413,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319431,8 +319431,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319446,8 +319446,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319461,8 +319461,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -319473,90 +319473,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -319566,8 +319566,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -321343,9 +321343,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -321355,7 +321355,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -321593,8 +321593,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -321604,26 +321604,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321634,11 +321634,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321649,11 +321649,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321664,11 +321664,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321679,11 +321679,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321694,11 +321694,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321709,11 +321709,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321727,8 +321727,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321742,8 +321742,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321757,8 +321757,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -321769,90 +321769,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -321862,8 +321862,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -323640,9 +323640,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -323652,7 +323652,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -323890,8 +323890,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -323901,26 +323901,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -323931,11 +323931,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -323946,11 +323946,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -323961,11 +323961,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -323976,11 +323976,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -323991,11 +323991,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -324006,11 +324006,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -324024,8 +324024,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -324039,8 +324039,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -324054,8 +324054,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -324066,90 +324066,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -324159,8 +324159,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -325937,9 +325937,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -325949,7 +325949,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -326187,8 +326187,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -326198,26 +326198,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326228,11 +326228,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326243,11 +326243,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326258,11 +326258,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326273,11 +326273,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326288,11 +326288,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326303,11 +326303,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326321,8 +326321,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326336,8 +326336,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326351,8 +326351,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -326363,90 +326363,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -326456,8 +326456,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -328231,9 +328231,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -328243,7 +328243,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -328481,8 +328481,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -328492,26 +328492,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328522,11 +328522,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328537,11 +328537,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328552,11 +328552,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328567,11 +328567,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328582,11 +328582,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328597,11 +328597,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328615,8 +328615,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328630,8 +328630,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328645,8 +328645,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -328657,90 +328657,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -328750,8 +328750,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -330521,9 +330521,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -330533,7 +330533,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -330771,8 +330771,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "lowered",
@@ -330782,26 +330782,26 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330812,11 +330812,11 @@
                 "sm_101a",
                 "ptx86"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330827,11 +330827,11 @@
                 "sm_101f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330842,11 +330842,11 @@
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330857,11 +330857,11 @@
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330872,11 +330872,11 @@
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330887,11 +330887,11 @@
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330905,8 +330905,8 @@
               "representation": "direct intrinsic probes below each paired target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330920,8 +330920,8 @@
               "representation": "direct intrinsic non-monotonic PTX negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330935,8 +330935,8 @@
               "representation": "direct intrinsic excluded-target negative",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -330947,90 +330947,90 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_100f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103a",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_103f",
                 "ptx88"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110a",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
                 "sm_110f",
                 "ptx90"
               ],
-              "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -331040,8 +331040,8 @@
               "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -332148,9 +332148,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -332160,7 +332160,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -332185,8 +332185,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -332196,11 +332196,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -332211,15 +332211,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -332409,9 +332409,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -332421,7 +332421,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -332446,8 +332446,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -332457,11 +332457,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -332472,15 +332472,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -332653,9 +332653,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -332665,7 +332665,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -332690,8 +332690,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -332701,11 +332701,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -332716,15 +332716,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -332904,9 +332904,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -332916,7 +332916,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -332941,8 +332941,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -332952,15 +332952,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "three convergent inline-PTX kernels covering the admitted tcgen05 commit and shift controls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -332970,8 +332970,8 @@
               "representation": "isolated direct-intrinsic probes at and outside the imported selection predicates",
               "stage": "backend_codegen",
               "mechanism": null,
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -332982,15 +332982,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing the three exact tcgen05 controls",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -333000,8 +333000,8 @@
               "representation": "the same three inline instructions emitted for excluded family and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -333233,9 +333233,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -333245,7 +333245,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -333270,8 +333270,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -333281,15 +333281,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "three convergent inline-PTX kernels covering the admitted tcgen05 commit and shift controls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -333299,8 +333299,8 @@
               "representation": "isolated direct-intrinsic probes at and outside the imported selection predicates",
               "stage": "backend_codegen",
               "mechanism": null,
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -333311,15 +333311,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing the three exact tcgen05 controls",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -333329,8 +333329,8 @@
               "representation": "the same three inline instructions emitted for excluded family and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -333590,9 +333590,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -333602,7 +333602,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -333627,8 +333627,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -333638,11 +333638,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -333656,8 +333656,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -333668,15 +333668,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -333686,8 +333686,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -333944,9 +333944,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -333956,7 +333956,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -333981,8 +333981,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -333992,11 +333992,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334010,8 +334010,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334022,15 +334022,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -334040,8 +334040,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -334269,9 +334269,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -334281,7 +334281,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -334306,8 +334306,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -334317,11 +334317,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334335,8 +334335,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334347,15 +334347,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -334365,8 +334365,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -334593,9 +334593,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -334605,7 +334605,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -334630,8 +334630,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -334641,11 +334641,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334659,8 +334659,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334671,15 +334671,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -334689,8 +334689,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -334920,9 +334920,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -334932,7 +334932,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -334957,8 +334957,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -334968,11 +334968,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334986,8 +334986,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -334998,15 +334998,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -335016,8 +335016,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -335246,9 +335246,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -335258,7 +335258,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -335283,8 +335283,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -335294,11 +335294,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -335312,8 +335312,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -335324,15 +335324,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -335342,8 +335342,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -335633,9 +335633,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -335645,7 +335645,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -335670,8 +335670,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -335681,11 +335681,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -335699,8 +335699,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -335711,15 +335711,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -335729,8 +335729,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -336019,9 +336019,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -336031,7 +336031,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -336056,8 +336056,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -336067,11 +336067,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -336085,8 +336085,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -336097,15 +336097,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -336115,8 +336115,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -336350,9 +336350,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -336362,7 +336362,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -336387,8 +336387,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -336398,11 +336398,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -336416,8 +336416,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -336428,15 +336428,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -336446,8 +336446,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -336680,9 +336680,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -336692,7 +336692,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -336717,8 +336717,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -336728,11 +336728,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -336746,8 +336746,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -336758,15 +336758,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -336776,8 +336776,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -337131,9 +337131,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -337143,7 +337143,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -337168,8 +337168,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -337179,11 +337179,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -337197,8 +337197,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -337209,15 +337209,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -337227,8 +337227,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -337581,9 +337581,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -337593,7 +337593,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -337618,8 +337618,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -337629,11 +337629,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -337647,8 +337647,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -337659,15 +337659,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -337677,8 +337677,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -337920,9 +337920,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -337932,7 +337932,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -337957,8 +337957,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -337968,11 +337968,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -337986,8 +337986,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -337998,15 +337998,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -338016,8 +338016,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -338258,9 +338258,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -338270,7 +338270,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -338295,8 +338295,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -338306,11 +338306,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -338324,8 +338324,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -338336,15 +338336,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -338354,8 +338354,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -338645,9 +338645,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -338657,7 +338657,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -338682,8 +338682,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -338693,11 +338693,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -338711,8 +338711,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -338723,15 +338723,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -338741,8 +338741,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -339031,9 +339031,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -339043,7 +339043,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -339068,8 +339068,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -339079,11 +339079,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -339097,8 +339097,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -339109,15 +339109,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -339127,8 +339127,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -339358,9 +339358,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -339370,7 +339370,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -339395,8 +339395,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -339406,11 +339406,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -339424,8 +339424,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -339436,15 +339436,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -339454,8 +339454,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -339684,9 +339684,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -339696,7 +339696,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -339721,8 +339721,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -339732,11 +339732,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -339750,8 +339750,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -339762,15 +339762,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -339780,8 +339780,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -340015,9 +340015,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -340027,7 +340027,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -340052,8 +340052,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -340063,11 +340063,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -340081,8 +340081,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -340093,15 +340093,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -340111,8 +340111,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -340345,9 +340345,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -340357,7 +340357,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -340382,8 +340382,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -340393,11 +340393,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -340411,8 +340411,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -340423,15 +340423,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -340441,8 +340441,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -340796,9 +340796,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -340808,7 +340808,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -340833,8 +340833,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -340844,11 +340844,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -340862,8 +340862,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -340874,15 +340874,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -340892,8 +340892,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -341246,9 +341246,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -341258,7 +341258,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -341283,8 +341283,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -341294,11 +341294,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -341312,8 +341312,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -341324,15 +341324,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -341342,8 +341342,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -341585,9 +341585,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -341597,7 +341597,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -341622,8 +341622,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -341633,11 +341633,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -341651,8 +341651,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -341663,15 +341663,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -341681,8 +341681,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -341923,9 +341923,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -341935,7 +341935,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -341960,8 +341960,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -341971,11 +341971,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -341989,8 +341989,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -342001,15 +342001,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -342019,8 +342019,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -342278,9 +342278,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -342290,7 +342290,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -342315,8 +342315,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -342326,11 +342326,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -342344,8 +342344,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -342356,15 +342356,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -342374,8 +342374,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -342632,9 +342632,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -342644,7 +342644,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -342669,8 +342669,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -342680,11 +342680,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -342698,8 +342698,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -342710,15 +342710,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -342728,8 +342728,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -343087,9 +343087,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -343099,7 +343099,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -343124,8 +343124,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -343135,15 +343135,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -343153,8 +343153,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -343165,15 +343165,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -343183,8 +343183,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -343544,9 +343544,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -343556,7 +343556,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -343581,8 +343581,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -343592,15 +343592,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -343610,8 +343610,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -343622,15 +343622,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -343640,8 +343640,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -343890,9 +343890,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -343902,7 +343902,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -343927,8 +343927,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -343938,15 +343938,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -343956,8 +343956,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -343968,15 +343968,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -343986,8 +343986,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -344235,9 +344235,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -344247,7 +344247,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -344272,8 +344272,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -344283,15 +344283,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -344301,8 +344301,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -344313,15 +344313,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -344331,8 +344331,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -344566,9 +344566,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -344578,7 +344578,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -344603,8 +344603,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -344614,15 +344614,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -344632,8 +344632,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -344644,15 +344644,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -344662,8 +344662,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -344896,9 +344896,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -344908,7 +344908,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -344933,8 +344933,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -344944,15 +344944,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -344962,8 +344962,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -344974,15 +344974,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -344992,8 +344992,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -345228,9 +345228,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -345240,7 +345240,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -345265,8 +345265,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -345276,15 +345276,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -345294,8 +345294,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -345306,15 +345306,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -345324,8 +345324,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -345559,9 +345559,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -345571,7 +345571,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -345596,8 +345596,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -345607,15 +345607,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -345625,8 +345625,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -345637,15 +345637,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -345655,8 +345655,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -345921,9 +345921,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -345933,7 +345933,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -345958,8 +345958,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -345969,15 +345969,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -345987,8 +345987,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -345999,15 +345999,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -346017,8 +346017,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -346282,9 +346282,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -346294,7 +346294,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -346319,8 +346319,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -346330,15 +346330,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -346348,8 +346348,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -346360,15 +346360,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -346378,8 +346378,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -346616,9 +346616,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -346628,7 +346628,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -346653,8 +346653,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -346664,15 +346664,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -346682,8 +346682,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -346694,15 +346694,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -346712,8 +346712,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -346949,9 +346949,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -346961,7 +346961,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -346986,8 +346986,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -346997,15 +346997,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -347015,8 +347015,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -347027,15 +347027,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -347045,8 +347045,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -347343,9 +347343,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -347355,7 +347355,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -347380,8 +347380,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -347391,15 +347391,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -347409,8 +347409,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -347421,15 +347421,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -347439,8 +347439,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -347736,9 +347736,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -347748,7 +347748,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -347773,8 +347773,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -347784,15 +347784,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -347802,8 +347802,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -347814,15 +347814,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -347832,8 +347832,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -348074,9 +348074,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -348086,7 +348086,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -348111,8 +348111,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -348122,15 +348122,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -348140,8 +348140,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -348152,15 +348152,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -348170,8 +348170,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -348411,9 +348411,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -348423,7 +348423,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -348448,8 +348448,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -348459,15 +348459,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
               "artifact_kind": null,
-              "tool_path": "llc",
-              "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-              "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+              "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+              "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+              "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
             },
             {
               "targets": [
@@ -348477,8 +348477,8 @@
               "representation": "the same paired fixture requested below the effective target floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -348489,15 +348489,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+              "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -348507,8 +348507,8 @@
               "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -348865,9 +348865,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -348877,7 +348877,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -348902,8 +348902,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -348913,11 +348913,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -348931,8 +348931,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -348943,15 +348943,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -348961,8 +348961,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -349315,9 +349315,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -349327,7 +349327,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -349352,8 +349352,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -349363,11 +349363,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -349381,8 +349381,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -349393,15 +349393,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -349411,8 +349411,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -349654,9 +349654,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -349666,7 +349666,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -349691,8 +349691,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -349702,11 +349702,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -349720,8 +349720,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -349732,15 +349732,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -349750,8 +349750,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -349992,9 +349992,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -350004,7 +350004,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -350029,8 +350029,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -350040,11 +350040,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -350058,8 +350058,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -350070,15 +350070,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -350088,8 +350088,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -350316,9 +350316,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -350328,7 +350328,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -350353,8 +350353,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -350364,11 +350364,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -350382,8 +350382,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -350394,15 +350394,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -350412,8 +350412,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -350639,9 +350639,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -350651,7 +350651,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -350676,8 +350676,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -350687,11 +350687,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -350705,8 +350705,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -350717,15 +350717,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -350735,8 +350735,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -350964,9 +350964,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -350976,7 +350976,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -351001,8 +351001,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -351012,11 +351012,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -351030,8 +351030,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -351042,15 +351042,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -351060,8 +351060,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -351288,9 +351288,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -351300,7 +351300,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -351325,8 +351325,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -351336,11 +351336,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -351354,8 +351354,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -351366,15 +351366,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -351384,8 +351384,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -351643,9 +351643,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -351655,7 +351655,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -351680,8 +351680,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -351691,11 +351691,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -351709,8 +351709,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -351721,15 +351721,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -351739,8 +351739,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -351997,9 +351997,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -352009,7 +352009,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -352034,8 +352034,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -352045,11 +352045,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -352063,8 +352063,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -352075,15 +352075,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -352093,8 +352093,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -352324,9 +352324,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -352336,7 +352336,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -352361,8 +352361,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -352372,11 +352372,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -352390,8 +352390,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -352402,15 +352402,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -352420,8 +352420,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -352650,9 +352650,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -352662,7 +352662,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -352687,8 +352687,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -352698,11 +352698,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -352716,8 +352716,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -352728,15 +352728,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -352746,8 +352746,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -353037,9 +353037,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -353049,7 +353049,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -353074,8 +353074,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -353085,11 +353085,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -353103,8 +353103,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -353115,15 +353115,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -353133,8 +353133,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -353423,9 +353423,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -353435,7 +353435,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -353460,8 +353460,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -353471,11 +353471,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -353489,8 +353489,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -353501,15 +353501,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -353519,8 +353519,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -353754,9 +353754,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -353766,7 +353766,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -353791,8 +353791,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -353802,11 +353802,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -353820,8 +353820,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -353832,15 +353832,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -353850,8 +353850,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -354084,9 +354084,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -354096,7 +354096,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -354121,8 +354121,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -354132,11 +354132,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -354150,8 +354150,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -354162,15 +354162,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -354180,8 +354180,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -354535,9 +354535,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -354547,7 +354547,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -354572,8 +354572,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -354583,11 +354583,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -354601,8 +354601,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -354613,15 +354613,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -354631,8 +354631,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -354985,9 +354985,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -354997,7 +354997,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -355022,8 +355022,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -355033,11 +355033,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -355051,8 +355051,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -355063,15 +355063,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -355081,8 +355081,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -355324,9 +355324,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -355336,7 +355336,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -355361,8 +355361,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -355372,11 +355372,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -355390,8 +355390,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -355402,15 +355402,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -355420,8 +355420,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -355662,9 +355662,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -355674,7 +355674,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -355699,8 +355699,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -355710,11 +355710,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -355728,8 +355728,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -355740,15 +355740,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -355758,8 +355758,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -355986,9 +355986,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -355998,7 +355998,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -356023,8 +356023,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -356034,11 +356034,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -356052,8 +356052,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -356064,15 +356064,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -356082,8 +356082,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -356309,9 +356309,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -356321,7 +356321,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -356346,8 +356346,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -356357,11 +356357,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -356375,8 +356375,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -356387,15 +356387,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -356405,8 +356405,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -356634,9 +356634,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -356646,7 +356646,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -356671,8 +356671,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -356682,11 +356682,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -356700,8 +356700,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -356712,15 +356712,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -356730,8 +356730,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -356958,9 +356958,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -356970,7 +356970,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -356995,8 +356995,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -357006,11 +357006,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -357024,8 +357024,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -357036,15 +357036,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -357054,8 +357054,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -357313,9 +357313,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -357325,7 +357325,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -357350,8 +357350,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -357361,11 +357361,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -357379,8 +357379,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -357391,15 +357391,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -357409,8 +357409,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -357667,9 +357667,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -357679,7 +357679,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -357704,8 +357704,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -357715,11 +357715,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -357733,8 +357733,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -357745,15 +357745,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -357763,8 +357763,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -357994,9 +357994,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -358006,7 +358006,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -358031,8 +358031,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -358042,11 +358042,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -358060,8 +358060,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -358072,15 +358072,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -358090,8 +358090,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -358320,9 +358320,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -358332,7 +358332,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -358357,8 +358357,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -358368,11 +358368,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -358386,8 +358386,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -358398,15 +358398,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -358416,8 +358416,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -358707,9 +358707,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -358719,7 +358719,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -358744,8 +358744,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -358755,11 +358755,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -358773,8 +358773,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -358785,15 +358785,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -358803,8 +358803,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -359093,9 +359093,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -359105,7 +359105,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -359130,8 +359130,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -359141,11 +359141,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -359159,8 +359159,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -359171,15 +359171,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -359189,8 +359189,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -359424,9 +359424,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -359436,7 +359436,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -359461,8 +359461,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -359472,11 +359472,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -359490,8 +359490,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -359502,15 +359502,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -359520,8 +359520,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -359754,9 +359754,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -359766,7 +359766,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -359791,8 +359791,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -359802,11 +359802,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -359820,8 +359820,8 @@
               "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -359832,15 +359832,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+              "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             },
             {
               "targets": [
@@ -359850,8 +359850,8 @@
               "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
               "artifact_kind": null,
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
               "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
@@ -360074,9 +360074,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -360086,7 +360086,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -360111,8 +360111,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -360122,11 +360122,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -360137,15 +360137,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+              "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -360352,9 +360352,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -360364,7 +360364,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -360417,8 +360417,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -360428,11 +360428,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -360443,15 +360443,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -360702,9 +360702,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -360714,7 +360714,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -360767,8 +360767,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -360778,11 +360778,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -360793,15 +360793,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -361050,9 +361050,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -361062,7 +361062,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -361115,8 +361115,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -361126,11 +361126,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -361141,15 +361141,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -361395,9 +361395,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -361407,7 +361407,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -361460,8 +361460,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -361471,11 +361471,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -361486,15 +361486,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -361737,9 +361737,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -361749,7 +361749,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -361802,8 +361802,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -361813,11 +361813,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -361828,15 +361828,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -362084,9 +362084,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -362096,7 +362096,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -362149,8 +362149,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -362160,11 +362160,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -362175,15 +362175,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -362434,9 +362434,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -362446,7 +362446,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -362499,8 +362499,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -362510,11 +362510,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -362525,15 +362525,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -362782,9 +362782,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -362794,7 +362794,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -362847,8 +362847,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -362858,11 +362858,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -362873,15 +362873,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -363124,9 +363124,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -363136,7 +363136,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -363189,8 +363189,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -363200,11 +363200,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -363215,15 +363215,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -363465,9 +363465,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_100a",
@@ -363477,7 +363477,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.6",
             "hardware": {
@@ -363526,8 +363526,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -363537,11 +363537,11 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "exact inline-PTX tensor-map swizzle-atomicity replacement probe",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained the exact instruction in PTX; the ID/PTX-hash list has SHA-256 588486907e9e70101b6445f23f07d27798a009766a831e519c21a0a2fe4ea178.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 02683e3a77ca0ae4f160645826b8e3da2b9da126376f94ce15265ad34e8ca282.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -363552,15 +363552,15 @@
                 "sm_100a",
                 "ptx86"
               ],
-              "representation": "LLVM-produced tensor-map swizzle-atomicity replacement PTX",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module; the ID/cubin-hash list has SHA-256 4bfa0ad94eac3615d96e64ef7594b7b7b96b549057da9dd83f38afd135c5aed1.",
+              "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9d36f1191de4ceed2adb6d065c793c92620591a4fbd3bafe95571a43268f4ec3.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -363802,9 +363802,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc-tma",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0-tma",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -363814,7 +363814,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-tma",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-tma",
           "target": {
             "minimum_ptx": "8.3",
             "hardware": {
@@ -363867,8 +363867,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -363878,11 +363878,11 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten exact inline-PTX tensor-map replacement probes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -363893,15 +363893,15 @@
                 "sm_90a",
                 "ptx83"
               ],
-              "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
+              "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
               "artifact_kind": "cubin",
-              "tool_path": "ptxas",
-              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-              "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+              "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -364104,9 +364104,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -364207,9 +364207,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -364310,9 +364310,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_70",
@@ -364403,9 +364403,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -364415,7 +364415,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -364428,8 +364428,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -364438,11 +364438,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "three typed llvm.nvvm.membar declarations with no arguments or results",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -364453,11 +364453,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "three typed LLVM probes retaining membar.cta, membar.gl, and membar.sys",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -364468,15 +364468,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 modules containing the three membar scopes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -364626,9 +364626,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -364638,7 +364638,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -364651,8 +364651,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -364661,11 +364661,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "three typed llvm.nvvm.membar declarations with no arguments or results",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -364676,11 +364676,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "three typed LLVM probes retaining membar.cta, membar.gl, and membar.sys",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -364691,15 +364691,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 modules containing the three membar scopes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -364855,9 +364855,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -364867,7 +364867,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -364880,8 +364880,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -364890,11 +364890,11 @@
               "targets": [
                 "sm_20"
               ],
-              "representation": "three typed llvm.nvvm.membar declarations with no arguments or results",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -364905,11 +364905,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "three typed LLVM probes retaining membar.cta, membar.gl, and membar.sys",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -364920,15 +364920,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX 3.2 modules containing the three membar scopes",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -365089,9 +365089,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-total-smem-size"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -365101,7 +365101,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "4.1",
             "hardware": {
@@ -365114,8 +365114,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -365125,11 +365125,11 @@
                 "sm_20",
                 "ptx41"
               ],
-              "representation": "pure inline PTX reads of dynamic and total shared-memory sizes",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted both size reads in 655-byte PTX with SHA-256 4ab2630ee33912457ed45c1cd516ae0797e062fcba7123db9bac352c67280b3a.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24cc1707db68cd33a711956ac50c8c65ab3e84eb5cf71a00fc15160d770b645d.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -365140,15 +365140,15 @@
                 "sm_75",
                 "ptx41"
               ],
-              "representation": "LLVM-produced PTX retaining both shared-memory size reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 ee3358b59d2b596f1b276d983c3149d58c384872ee5edad482cf20a99183e7b1.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7feb7ce1442470cb7ca3f613338c9fa901e16b23bc621a84ac3aa46a8583088d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx41, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -365280,9 +365280,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-trap"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -365292,7 +365292,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-debug-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-debug-control",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -365305,8 +365305,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -365316,11 +365316,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "one LLVM IR probe with side-effect inline PTX for trap, brkpt, and pmevent 15",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -365331,15 +365331,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "1,011-byte LLVM-produced PTX module containing trap, brkpt, and pmevent 15",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -365507,9 +365507,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "lowered",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_30",
@@ -365616,9 +365616,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-warpid"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_20",
@@ -365628,7 +365628,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "inline_ptx",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0",
           "target": {
             "minimum_ptx": "3.2",
             "hardware": {
@@ -365641,8 +365641,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -365652,11 +365652,11 @@
                 "sm_20",
                 "ptx32"
               ],
-              "representation": "side-effecting inline PTX reads of smid and warpid",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "Pinned llc emitted both volatile location samples in 633-byte PTX with SHA-256 8b6bdea0bf90c68baecd7189ae84ba5c24a0ca32a3a347a1dc72ca8a0a62277f.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ec007869a74a8d4d9bf7be114f276c2c4e26f3e4c34eeaab721362bbf9a1918.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -365667,15 +365667,15 @@
                 "sm_75",
                 "ptx32"
               ],
-              "representation": "LLVM-produced PTX retaining smid and warpid reads",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas produced a 2,984-byte cubin with SHA-256 2a1f9febb72d01384c5c1c4583e4511020b2c086144d8b25f6aad354a4c34618.",
+              "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14f47ee4cc2165c26a56b268fb647e325de01ae94e284257de734b97dea0626a. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -365834,9 +365834,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -365846,7 +365846,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-wgmma-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-wgmma-control",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -365859,8 +365859,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -365869,11 +365869,11 @@
               "targets": [
                 "sm_90a"
               ],
-              "representation": "typed declarations for all three WGMMA controls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -365884,11 +365884,11 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "one typed LLVM probe containing all three WGMMA control calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -365899,15 +365899,15 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "typed LLVM PTX retaining all three WGMMA control instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -366088,9 +366088,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -366100,7 +366100,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-wgmma-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-wgmma-control",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -366113,8 +366113,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -366123,11 +366123,11 @@
               "targets": [
                 "sm_90a"
               ],
-              "representation": "typed declarations for all three WGMMA controls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -366138,11 +366138,11 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "one typed LLVM probe containing all three WGMMA control calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -366153,15 +366153,15 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "typed LLVM PTX retaining all three WGMMA control instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
@@ -366349,9 +366349,9 @@
         "ptx_isa_url": "https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions"
       },
       "backend": {
-        "profile": "rust-llvm-23.1.0-16696adc",
-        "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-        "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+        "profile": "rust-llvm-23.1.3-aab11ee0",
+        "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+        "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
         "status": "validated",
         "target_triple": "nvptx64-nvidia-cuda",
         "gpu_target": "sm_90a",
@@ -366361,7 +366361,7 @@
         {
           "backend": "llvm_nvptx",
           "mechanism": "typed_nvvm",
-          "evidence_profile": "rust-llvm-23.1.0-16696adc-wgmma-control",
+          "evidence_profile": "rust-llvm-23.1.3-aab11ee0-wgmma-control",
           "target": {
             "minimum_ptx": "8.0",
             "hardware": {
@@ -366374,8 +366374,8 @@
               ]
             }
           },
-          "version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+          "version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
           "artifact_path": null,
           "build_id_prefix": null,
           "status": "validated",
@@ -366384,11 +366384,11 @@
               "targets": [
                 "sm_90a"
               ],
-              "representation": "typed declarations for all three WGMMA controls",
+              "representation": "separate typed LLVM declaration probes",
               "stage": "declaration_canonicalization",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -366399,11 +366399,11 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "one typed LLVM probe containing all three WGMMA control calls",
+              "representation": "separate probes for the recorded lowering mechanism",
               "stage": "backend_codegen",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3.",
               "artifact_kind": null,
               "tool_path": null,
               "tool_version": null,
@@ -366414,15 +366414,15 @@
                 "sm_90a",
                 "ptx80"
               ],
-              "representation": "typed LLVM PTX retaining all three WGMMA control instructions",
+              "representation": "separate PTX outputs assembled for the recorded target",
               "stage": "ptx_assembly",
               "mechanism": "typed_nvvm",
               "outcome": "succeeded",
-              "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3.",
+              "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8.",
               "artifact_kind": "cubin",
               "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-              "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-              "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+              "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+              "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
             }
           ]
         },
diff --git a/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/README.md b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/README.md
new file mode 100644
index 0000000000..84a549b5b5
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/README.md
@@ -0,0 +1,133 @@
+This directory records the 2026-10-11 LLVM evidence refresh. It is an audit
+attachment; the evidence scanner reads only JSON files directly in the parent
+directory.
+
+All **1,699 primary captures across 1,029 intrinsic routes passed**, including
+1,506 terminal assemblies. The initial candidate sweep had 192 failures: 148
+renderer panics from incomplete candidate records and 44 invalid combined
+LLVM target/PTX requests. These were rerun using the existing full-catalog
+renderer and the original evidence's separate compiler and assembler targets.
+The final captures include **56 cases with different codegen and terminal SMs**:
+the 44 invalid requests and 12 cases initially blocked by the renderer. For
+example, LLVM `sm_20/+ptx32` output is assembled with `ptxas -arch=sm_75`;
+the capture does not claim LLVM supports `sm_75/+ptx32`.
+
+All **377 auxiliary captures passed**: 187 typed declarations, 116 typed codegen
+checks, 66 typed assemblies, and four inline FP8 alternatives checked through
+both codegen and assembly. Compilation checks used the existing
+`probe.rs::assert_intrinsic_declaration_canonicalizes` and
+`probe.rs::validate_probe_instructions` validators, including imported
+declaration attributes and exact PTX operand/selector checks. Historical
+negative experiments and hardware execution were **not rerun**. Their old
+compiler provenance and original outcomes remain explicit in the parent
+evidence files. All 38 historical stages have outcome `not_revalidated`;
+they establish neither rejection nor execution with the refreshed compiler.
+The failed typed-route admission checks apply only to unchanged libNVVM
+evidence. The LLVM historical negatives do not authorize current routes.
+
+The tables preserve 770 refreshed stage groups and 3,753 artifact proofs:
+
+- `metadata.json` defines the columns, zero-based indexes, exact tools,
+  original paths, source identities, and command argument templates.
+- `captures.jsonl` contains one row per primary or auxiliary capture. A null
+  artifact slot means that capture did not produce that artifact.
+- `artifacts.jsonl` deduplicates SHA-256 hashes and byte counts.
+- `stages.jsonl` maps the new evidence filename and JSON pointer to capture
+  artifacts and the aggregate ledger digest cited by the evidence detail.
+- `inputs.jsonl` retains the exact UTF-8 source for all 307 distinct auxiliary
+  LLVM inputs. No temporary helper binary is needed to recreate these inputs.
+
+To reproduce a capture, use the recorded tool versions and argument arrays in
+`metadata.json`. Primary sources come from the existing
+`render::render_probe` implementation in
+`crates/cuda-intrinsics-gen/src/render/probes.rs`, through the full catalog;
+the generated files are `intrinsics/probes/{id}.ll` or
+`intrinsics/probes/{id}.llvm_nvptx.ll`. Every captured primary source was
+reconstructed byte-for-byte from these files by restoring its original
+catalog-hash and LLVM-revision comments. Those original comments identify
+the inputs at capture time, before the evidence refresh changed catalog
+provenance. They do not identify the final catalog.
+
+For primary captures, restore the catalog hash from capture column 7 and the
+LLVM revision from `primary_header_llvm_revisions`. For auxiliary captures,
+find the source in `inputs.jsonl` by its LLVM-IR artifact index. The source's
+hash and byte count must then match `artifacts.jsonl`. Derive the original
+input and output filenames from `original_paths`; the column definitions
+name the values substituted into those path rules. Run the applicable
+commands, using `codegen_gpu` for LLVM and `terminal_gpu` for ptxas:
+
+```sh
+"$LLVM_BIN/llvm-as" "$INPUT" -o "$BITCODE"
+"$LLVM_BIN/llvm-dis" "$BITCODE" -o "$CANONICAL_IR"
+"$LLVM_BIN/llc" "$INPUT" -march=nvptx64 -mcpu="$CODEGEN_GPU" \
+  -mattr="$PTX_FEATURE" -o "$PTX"
+"$PTXAS" -arch="$TERMINAL_GPU" "$PTX" -o "$CUBIN"
+```
+
+Run the first two commands only when the capture records canonicalization,
+and run ptxas only when a terminal GPU is recorded. Declaration-only auxiliary
+captures stop after canonicalization. The original source and bitcode paths
+are significant: llvm-as supplies `source_filename`, and llvm-dis prints the
+bitcode path as `ModuleID`. Restoring those paths allows byte-for-byte hash
+comparison; other paths provide equivalent compilation with different
+canonical artifact hashes. No `source_filename` override or path-remapping
+flag was used. The normal contributor gates remain:
+
+```sh
+cargo run -p cuda-intrinsics-gen -- check
+cargo run -p cuda-intrinsics-gen -- probe --all --skip-terminal --per-target
+cargo run -p cuda-intrinsics-gen -- check-abi-history --base-ref "$BASE_REF"
+```
+
+Extraction used public LLVM revision
+`aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019`, Git tree
+`743426f16c6d14d89426972b8aad1540e5d25ab0`, consumed by Rust commit
+`6f5da406b56c2db6607202b63daee73aa8c61297`. Build `llvm-tblgen` from that
+revision, then extract from the CUDAoxide workspace:
+
+```sh
+git clone https://github.com/rust-lang/llvm-project.git "$LLVM_SRC"
+git -C "$LLVM_SRC" checkout aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019
+cmake -S "$LLVM_SRC/llvm" -B "$LLVM_BUILD" -G 'Unix Makefiles' \
+  -DCMAKE_BUILD_TYPE=Release -DLLVM_TARGETS_TO_BUILD=NVPTX \
+  -DLLVM_ENABLE_PROJECTS= -DLLVM_INCLUDE_TESTS=OFF \
+  -DLLVM_INCLUDE_BENCHMARKS=OFF -DLLVM_INCLUDE_EXAMPLES=OFF
+cmake --build "$LLVM_BUILD" --target llvm-tblgen -j8
+cargo run -p cuda-intrinsics-gen -- extract --llvm-src "$LLVM_SRC" \
+  --llvm-tblgen "$LLVM_BUILD/bin/llvm-tblgen"
+```
+
+Both TableGen dump hashes match the previous extraction. The normalized
+2,633 declarations retain their semantic facts; the normalized-file hash
+changes with source provenance. Exact dump commands, hashes, the local
+TableGen binary identity, and the independent LLVM 22 intrinsic-dump
+comparison identity are in `metadata.json` and `../../upstream.lock`.
+The public clone above is a reproduction alternative; capture used a local
+shared checkout followed by a depth-one fetch of the public revision. Its
+exact checkout paths and command arguments are also recorded in the metadata.
+The local TableGen binary hash identifies that build; matching dump hashes
+provide the reproducibility check.
+
+The following check reconstructs every stage digest without temporary
+artifacts or helper programs. Run it from this directory:
+
+```python
+import hashlib, json
+from pathlib import Path
+
+def rows(name):
+    return [json.loads(line) for line in Path(name).read_text().splitlines()]
+
+metadata = json.loads(Path("metadata.json").read_text())
+captures = rows("captures.jsonl")
+artifacts = rows("artifacts.jsonl")
+for _, _, _, expected, proofs in rows("stages.jsonl"):
+    entries = []
+    for capture_index, kind_index in proofs:
+        capture = captures[capture_index]
+        digest = artifacts[capture[6][kind_index]][0]
+        entries.append((capture[0], metadata["artifact_kinds"][kind_index], digest))
+    ledger = "".join(f"{name} {kind} {digest}\n" for name, kind, digest in sorted(entries))
+    assert hashlib.sha256(ledger.encode()).hexdigest() == expected
+print("Verified all 770 stage ledgers")
+```
diff --git a/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/artifacts.jsonl b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/artifacts.jsonl
new file mode 100644
index 0000000000..3c08b643ef
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/artifacts.jsonl
@@ -0,0 +1,4899 @@
+["b5e4853267c0f13a342903b805930375918f6cdc20a09d160e61fdff8521cce4",555]
+["7fa85a6b7393844c634504c3dd2dbd922732d9acbc63d628bbb0aa4768fc6562",568]
+["597b86955db03bad4e230701c2144c1790aa549cb78dd554ffe7316794b4045e",1464]
+["5ada19d90c04fae5fa6b4d25e22dbae6dae1d20986b08508e1f856167ad2ff67",553]
+["28a75746e0ece23704cdef5ad42698f95420fc2bcda59499adce3613b56a5367",562]
+["532697b7b8301fb80c772d62c8ada76415ed54bc2096b274be0e766b0831fb4b",553]
+["553e6381b901157f4085dfe8001caeebea107115d4856254d0b20ed6b9e6dbea",562]
+["afbdd38c63b7842c55bb0ba5902640be1b322162fa350de10a3d4eee16cda54a",1464]
+["7ac3a7ad60d79d7d4e97608c61d009ea005b79b395e8993e0eeb7f9b7d863ba9",565]
+["fbedd80e5a07a45faed279d2fa13cc9cd10fb367746b2faf194ecc86c3e4e15a",2300]
+["3f9443969fe05823e7a726bfa52dca9aedb7877f8b02884d86aac0c2d9be7787",576]
+["3446b2bfd0b7859afd55e1aa7602e4decdf2048e7bf0078a6ed722a4babfae11",444]
+["858d72f31eeeb904cc45696c9b420b11c99f5f8c6873f39285b64f0930a502aa",586]
+["deff26205bfaa83c852046f2c5039dd05d2cd241cf853d8e7a9cfe94d8e6da0b",663]
+["36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c",1720]
+["ea5f53ddd31cc0d552d891922677e6001b98b12f9a1f17e3e02a9afd43fb597b",584]
+["c2f24061b4b9ffea6f080940ba855b86cbb6a56197b0ef7b2a88e588f95f343a",655]
+["aa7e4c769bf11d192a3aa381db177785d23c3a979d626673d52df94ccbbdcdef",584]
+["cc4c1199eb4c1db5fd254755ed7463b1305b93f6faf2de9c9f2108c02d1435cd",1464]
+["264d5a4821ae6536e422cc6b2ad93e2ecb3a843454ae6cddde542766602896f0",584]
+["33e2f2fa2d887fd9e1acdb8925f871a3cf92d027e80468ef9c3b977cf6f3e919",663]
+["3b98f5284574d39beb0194db60153831911e5733920bb6badbf9a3b3cf846cfc",2240]
+["5c49cc0baab7fca5fe9e71e3dedf2dea709b126ac6e453fc3ec4245dc2e4c4ee",592]
+["4e6d49d69a0ce4db1fce0e94790cb853df773b9b22d5a9aa8d4d4c4beee70ab1",695]
+["6263b146fedc87f4d49df3e0cbbd50d747b8ba74a43fc5e91656e1c03b80f003",627]
+["7fa9c8c62c07fce2dc6c443b2d576b7b70851e8e7eff6ea5bf3eb048cc311e14",725]
+["308b9ec5646e1f3299784c602b5d09430e243eaa2054a2dc61c963cf19701fe3",621]
+["cca642abe24116d3ab2f90a9939b227d164c978be119a859f8f1cdf7e172a919",2344]
+["5235b48c6c5dcb264a4dd1cbd2773b0209a2470ae31111dc68d506c6ab3cfd0c",637]
+["a7d1f99b8b78837d900e74eb5ca6ce5da47f41d4bc7fe4366420794f8924bda7",735]
+["648cb79ee87f619c77a7054be7cee364443c411573628693867ce4fae03d13a8",628]
+["05d57982746d69a5096fa4fb96975870ba6d49c3c5ffa3690631fa0f366513cb",2344]
+["fafc5980f75d9b8394b4be23f677c1abd14470021254f2711a94b0a9768ee083",639]
+["071b76a4e8d6c0c463a42dae55a5da01e4ffd4d0206b229dc2bf3e61643921de",749]
+["021d0e11f13ee02e34772d8f0b65a0d8b5119f8e398ae4363aeb78c2b0cedbd5",653]
+["4b0f42b21620ac7730ece5c4561cbfebb1e1ec4ca2c73778fc9eea2f58b84917",2360]
+["c17a912405ffa810f12f140b694262bcead38f037f4be2a25ffd223316958c02",613]
+["756c976b032191c9a014bcb5b8e0a8dc957c3261bf6c09630175b41207c36720",725]
+["efc35b75806b4c57c5d48891ba3572d39e53019bff9bd1fbc4168d142048f67d",605]
+["ecbd1c9d1aa7ad61e4acf7c71aa5d6e0852c0acff56a28c93684a0cc170dadbd",692]
+["071d4f2cb125b7eb6b8c6ef4c51b06849785955b9e39bd4cf524690927d2f032",627]
+["5e8a375ebcb7ba5ee35fcc88058105ae9e5165f467948eef3753914c85637716",725]
+["c93c2d493888c20e0b170c40eb3f45a364360234a766b77e1b859bef63146cce",621]
+["c3c11e0970ef2baccf00bb66640e7358161dba3a28564621aa13baaef480c2b0",2344]
+["92f101a35eade06cef57831ca2ebb1392a85430c8ab66427e4018ed03ec4b10e",637]
+["3c99c448217cdc5fbfc35019015ab0d4b81fefb5787dd5d71ef30b734e6d33f9",735]
+["59874b0ca2d243819d526c51d82e6df6aa73c2d7c9d70ef99128a0c6ac0c1cb2",628]
+["71388b900425130cdd88d99bbaee471b7459f0c6ce8c1af02daa51278225aa6f",2344]
+["e243b5a1cd0484b648c0b88df7856d2bf3e2e372f7b3486612e2dfa48482b45f",639]
+["f2f6d8adfcf63ab0b521b162b0beb25c98ec3242c849cd79a95e144f69cacb79",749]
+["d4069882a72e5f89da863bb8bae37b7097deee98ffa5de040baaf159c903bf26",653]
+["cbee66c983ee082e1970965db4aa62be06225bd744019edda7f0bc02daa9eba6",2360]
+["9d19680f7311501c1202bef675228efff05db051c84122b432828a2666dbe2b3",613]
+["924edbe2ffaae8fa3816a058a7f950571a69c7ee300408b0fb8e7551060b1e41",725]
+["0cb30af394263adf2ac7f16e492b761850bb1d760a08c066d72c701bdb8fdc96",605]
+["5562bb5adffc4c40bab6f5bc7351a9253d413a31993119abe5bdb8498f3bb1b5",692]
+["22e95a10989a1b9fb1893f3ac1309d23827a06595568094fc4c1b2078913b741",627]
+["a7831b13f0b29a46c287928f7ae7b540dc8090ff490cca8eb30d7e63a8f6b26d",725]
+["0a68e3b676dd7587ae2cb2c16dd9a9656dfcefe68c7376a44a4d152878480062",621]
+["57738c12a9a95b85135b8988f58a238293ebef5f97cfdcf668a1e90be75d3d15",2344]
+["38c415fb4291d026af53bbe843888b2957e95d1910bcd9672a793cd971073e2a",637]
+["dd1960aafd2e54df6b6f5307547114f9f85e75ce9b730b9b395c172eaab1a520",735]
+["3486bd469e88e5052277f1a8b774a3c1a429c199d11dcd47c08975b392d12679",628]
+["f43027771ecb6b784655dd51ef971cb7b001e5c420fc93d479784f61932f05bc",2344]
+["6ed4dc652c7259e9ce3bfd123ac6d21fb17884a0e143a44d29c6aa9846432b5b",639]
+["4525cc1f5bd39441c484fb7cbd4b390cc895400a3826a89da339549f19c64cc6",749]
+["e3a60873cea95150fefb1523575683add85cda1ee4c580331fefcb785c66145c",653]
+["8c846876305928554f37897dca4e4c813d38ace8b725a02971c80206c79f0c79",2360]
+["ef245e8f402f2ec3747b234ee2b0147deab8211e7994d7bb0a0a29b10aff483e",613]
+["8885387a4d1a00a338dd8ae5a96e3ee43a099814eef37525bc39d333cb054621",725]
+["bed8ae62970231da5dd9af4889b2a516391de4d390fac71da05fd49b23c02aca",605]
+["587893b74263834bd36ddc4b78aa100d4afa552b9c0af88ff7ef8a674c9f9504",692]
+["9cf14c1e75d88ea7ae6fce2cbcd57537252a8e253844fa94397f5c5216e8d457",627]
+["ccc2a6c371db133edc03be9a654948fdc81f14dc8784846288bdbab4bebcab51",725]
+["f26def04381a6a3a75cf60b1e0b9d7f37efeffe57ecb884050bb73bc6cad8377",621]
+["34112e5e295bc4397d407932ca1f8a46906f9b103d2adfd5444c255cc439530c",2344]
+["eb9ba440ec7e277da55e7a3846a75473a75ac9008088fd50c87070f364c660c8",637]
+["1fa6d092f97c7c071d6bf765b9d853461c83230a371fa6c8ac1cf60d930a71f1",735]
+["417f194cfc33791ba06a4c10ac95340ab6468989611594ccc46938fd21f7dee2",628]
+["cc007227d93c8d8d96cd581011ea90919ecebd0b96f5286705bcf0c0a46ffb0f",2344]
+["fd8b4a8641cdd5bd7c8ff76772df094fefd69b55e6ef5e090453913f3d9b5bb1",639]
+["c9774dc9f1ca38c48ab7b5f3dc470cfc9bf2c257be2e26a4fcc6ae51b68ceeae",749]
+["a538acb5770c6219d78df5d4263cd509b0eaccd3c0febf267c170221d2999633",653]
+["b243b2199e798b537234b1a027dc3432a0e68a3d5657d5eaad3734829572ebfa",2360]
+["6e5591adb35b43286f29368b7619021a85ab527feb743e93a9a8c3c87bac938b",613]
+["bc34e6371051a851068e5e4f4b7c72a9f0ce492a86ab21e54ff29052f545e46d",725]
+["b4cd2a329ad82c6831f24e48a529e0a10a30264fefebd05253e84f1128895f54",605]
+["dd89eabfd5fa32429580924774cbd5fb7e5be56f8d524c40986fdbaa31d94587",692]
+["aa7646232d39e0146212587671d0e27b637d270a90a1b4d69b7b1782a017642f",772]
+["b5443070446b3b0f25854d48fd7b6f4e86c34fd0f2716e373626af2289c10946",2440]
+["3301c7d7c32a644230d8c66544217156173f60c6c61a6bb499e70cb25d665092",779]
+["e57be908c5881b03bce4a16bde1fa3c3e0e0861948b5bb9686ea460def7e2cfb",1349]
+["c16d4a8146bdef933d7d7234665c0c2839e879195f791a0a3b88e8a0ab4b045f",772]
+["a0b92312ae186824b89d5578ff4a6072a93a3059e7c51cdf4f41f156de2dcd64",2440]
+["2cfad5e49de345b0ba2ac41268ecf5229e7f51acd400ef4443325fb5fd3882c0",779]
+["1667fcd98c4d0b0d8cd764d8b33d0e38071c5c1f37948af91493414484d1ab95",1349]
+["5a81d9009a5d1bf28e00d58af6cd4520c3fb4b05b3765527ab89aaf5aa24b26d",794]
+["d352f2d9dfd5b95a1640c4dbc5954fc4b54ca9edff2bbe24650800d2298747cc",2456]
+["1662fcec88597c585961641aadbe3d722176924aa1009f544b9f37bef5051757",807]
+["4aae99b651e3984c18199eb40b22ef8f9d3dadb1331fdb6221fb6362bbfe5144",1336]
+["bacd4cca08140c3636b420e071333b63577049f2c8c2195e5f52e773e591cbfa",626]
+["04c72e3e44c19b82b5640594f4b74318e8c6de47b191451e58eb2387687ac1e7",2336]
+["bf66b85b067c7c937f7ab7135434bc4de19dc500fcd7eb2c8a2c45df8b8e0781",615]
+["d68186837e26f2fe37292bddfcdd299c010635eefd75cf51feca01a16f390c31",403]
+["a0981b3a316b510f1686a4f9da8434b0b118917254864750a692d29a009cfa39",626]
+["bf37f22c6941add930f49a7f4b0120903cf924dea8f4dd403b9bfd678166d0a2",403]
+["bf6c12f24303c0f9931dc588981252511083d6792a176276c17b797d7f934818",650]
+["c7f7dce25d7dc80925fc59a7a51ce1682efdae0752bcc30fe476ef4c3f1d1009",2364]
+["696d70b4e6ea9ebf288230eab815f422c214bf2912b88c96113f1a66425cf67e",655]
+["a940425bc69d88e2d08655bbd224c63fc6a9c5482dda967a7baf101b773da29a",443]
+["6e203d36b8688754357f7473e2a852ad7435d2ea533fc04df60725e4b712e20f",650]
+["42434a9d9438978cccd3e686adeee8a39e0a982e53229522ff6c3c9b8cd71d56",443]
+["1193f8f843f036a098e00fe67921fe3b7ffa01832a8c135b09bf76d9c7274ba3",650]
+["f1b064e4e2847ea1b8f7eb0b71dba0ca51c0d7af2f2dee0eff1a059dc3c425f9",2364]
+["925e0d55241330ae2984cfc96d6db31df451d8d6e95964a53f9aebfa281cf65d",655]
+["b65312e1e143077d2129d553de4ff8dcf265a2210af49b290b4463d265c58134",443]
+["ed6232e2eaac6d6a717e0be29277858f318c60e0a0238dc73194152aa3dfe27c",674]
+["b65ddecf1ea59edbc26a1fcd7045b7cc0c6850ad13de4b1cc4f64799da1e5359",2396]
+["ff87d4b2650cd6f3e4320c0bb3efe59dea3dc9094ac341c5dfbede995341553f",695]
+["5a499617767c673aa857fa146e44f02c08ac2d60407ee9723b1fcf2dbf68bd11",483]
+["81ecaf5b50e456d3d117db6a1cb9a9c768e029945d968bc39740500d7cd71120",620]
+["999a218f18c9cc180eba3a46e94174ccca817becd9714092427e934c235c3838",2320]
+["ea740fb79a59c3c6d5deb4a67ea2edb57912273a5434ff3596d73bebc763e965",605]
+["6aae19319dcbab33203dc9870f4e287dcbef48a2d0f3c960ace8164da7063650",393]
+["d421c15df6e436d7d7247536c5374076509051f81b1c391ef883c136480b65a6",620]
+["4744f95eba73cf72775aa83795bb179f76091e1d134243095ef41754d3507b14",393]
+["d5a6a322cc4a0d261ef2a8ec1701875e28b79ae36f4ed352da4a1392a3be52a0",644]
+["4743aef697d06decab81c2a21f5236a2dc71d493d392b4e604863bc3e4ee2d17",2356]
+["dfe95470506c8154d68847a157ee10ab44574747abe18441b86e3f1c3205e87c",645]
+["fb1f451604ef79c43c3cfdc4c56f0067732b744d12f8c56d48a7a661d6d214d9",433]
+["24b9e07c8a78d58481a68c2a0768937009f640875eb3591f4b64b59e6846e2e7",644]
+["af6dd86716ebc62a3d19dffa6dcfcb17cbfc0d47ed80f456a436b5a13d865061",433]
+["638d10a88f59aa0274c6850698cc705e6007d044374bce176b026a4f0355f20f",1127]
+["76e8e4e7fc4d0da1f32aa21bba32fead911ade64373219643f524bda2158098f",2700]
+["93b80970e770ee7991104964cde5ce1ce4536539a2022c4fe473eb7b702d0353",1198]
+["94113e81da1085d1688c37f811dc497e0963be3af0f703f03862268e4a14678f",1780]
+["491e591c2aef5f2bbcb89330ca165c238634325b11e653d923dbab28cedd13e4",1127]
+["5814251a5d1e180fff679013f344f9a17f04a0e3266ecef12134e6703128f41b",1240]
+["ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857",1464]
+["f386efa943732fea7b5d77d10a61c2b7a7349ab65c683a5f8eb32d0f368fd826",2728]
+["949c31afa15c65ecc7a881483a525489d504357cf7d7018dfc3c8c628d77609d",1199]
+["9a56dad3ad1c0c574e4f9d1b20f6afe7cb7248a9de3423043467ca8e706f6a5d",2756]
+["2193d71bd9bf8b2868610ae72d1a47b25ade23cb243216bdff794e6e76f2815f",1286]
+["6b4d17807a12f8298fca5f4957b8e3da5ed27fcbd47111f8a173bbe61d5c8aad",1956]
+["6d2b1a894de161307a6b17c33579ea9fb061c13c5e7a84b23c6cca6f370686ab",1199]
+["05b8df8bcab668f9e7342396f76ea79803a458885e664534c754cb24ec68c7ef",1336]
+["64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05",1464]
+["ea7dac3c6e8a3b347340985ae980f02ba983769b70ea3c7825171028f06caa40",2784]
+["ba37f1cc9be8b2426d2b7cfb449f01ea33831acc7ec207fc8554a9b13b44f1b7",1109]
+["3567b8a103f48bcf39f324e228bfc01b7ed5afe16188088c5f76d72cf907204a",2684]
+["22e8754719525eec7838c9788920e13e187860752581c1058a9a7a8f76efb647",1176]
+["b220ca5fdae82f6048cc1e930fa07a9f6e8cf23a610476f0cf469a01e985a85d",1724]
+["cf32b135bd8d84156d0e6a17f52c27528300ee7af2b0df7d465b69d7ff5989f6",1109]
+["97fd1fe4e29fd8f135dc5da2b3c86ba6c2dc671b429214ff3ba7c56d76eab744",1216]
+["6c57421050876bf374d83632bc8e985b7a17b643e44091e892350d3d005e97bf",2712]
+["72353df2cdcf40d63f84da33aab2b271fddc51dec1debe72e25e901947f6dfab",1181]
+["c508d3e05c68f3f4a7cd68936b10d60b533ce1e663eb8738baf2bf1496bd7dae",2744]
+["15a6df475009cd11bc0c19873d8b2bbcd288031471bedba8c310e87c0a421767",1264]
+["d3fbf9d7aa025ee2f4836094b8652fb29a5557e94bbc73689c5b18dd04edc47a",1900]
+["107cde694fb4a9a30b3f5005adef66b6fe55cfc08b4e1ca8c1381da65ba8cc09",1181]
+["39747bbacbce8a9980582496efe2b7103470e381f832d4cfb7bd3febc9bde506",1312]
+["962511241e3844fb4afa54c3cded2b256d986114bba593832da3cee47a46e5ad",2772]
+["844e49e0baa01f7f4fc32a30f0823776084dd90e3a1b3143f0dca99c1957ad11",585]
+["fbdb98ebeae05f8011a13391188ef92d85eed5c79c92e37d350ab5b3716f07c3",2332]
+["26dbbf0c2190c536f34a7e0f60e62f2f7f0b5650b81a9d9d088b639f8678ccf2",633]
+["7ac8d02201e099f2746a2660a91d4f9d89950dd88703e79ff6c20d4507c05ede",446]
+["67f17768a0923935794046905c50d89bedbcb18dcff743885a6e8f84499d444f",585]
+["ed1b37fce7f2c7ad65a49752e5e1a8f5009f1c343a52054bb5eb8f4baee9636d",2332]
+["69f5a36714c81ed80100895435b864cf672508b8c35df0b8f5d6876e3578f70a",633]
+["0d6079cd3cc868a57134aeba0eac250ffcf1b6780ff9cb0a59687fdc1f55bcd2",446]
+["e201c5a70c205ab0fc5b3c8efd657bb8549dad61ba40e4a89662541c3bb3ad84",585]
+["84e4da3851efe301345a98ec4091f6308f5508065d7f205cd99b267cb1886162",2328]
+["83ea328e35c42babe5605ee75e367c0c6f6676dcf6ec5f123dfa80a121efb3da",631]
+["2c22c73000127d7227ac9e7ed9b1a8b6b19f0fb8777ad10eeb77d1360a18065a",446]
+["851fac64b42ecb81afd9f84e606d5442e4160563a79b829b09d429bd16e5ae2f",587]
+["7f1755a0044c1f3224b1e6f96a9ae65521af684ca6a1d28cbf0c42c2d5bacc69",2332]
+["311065b79d079dfaf9871ea91a7d878cbdf79262475250f90db3b7fd9dfcfc21",641]
+["1bce82a3ad1ba7509b2dd219c8b0bebe93adebdc217aec6f0ffdcd38f6f4ccb7",447]
+["577f39d0630638f5faae09b885abe2cb029c2d899feed3b0b8ff27eebcfe3847",587]
+["bcc394433a1b1fe3b55263b73918e6fd5fd308002e9bfc671e413e9c81bd5021",2332]
+["59e9ffcd08e708d13f1f3a91b0d88567430b1d312212b3444910508cf1a8dddd",636]
+["7f568679dff10fabb11daed666e46d3dec4d2003f7737a3b0df3d968fd12ea5d",447]
+["3c7699b9454e1c7c21c94d3ee59439690c9214828f289fb193eb5498acc213ce",587]
+["b2163f638113897cb6a081c6acc42f1cc71ae04ece907a3ba66481ed740bbfd2",2332]
+["a97eb1d0c73335465797977dcb6b717e62e1d6819f69c73b265eb72a50b8e42d",636]
+["6101b80b18a59789b84a6c1a4fe893d38fb339570809e772a86fc422c6b236df",447]
+["24055778e2598e85d48e12fc739332e99dcb48db624de91948189a11671a38d8",517]
+["39e8df89aad14ef479c4923d0166b3e8565ec0196558143cfcc88a9f3b28e9a4",385]
+["08ecfff730ab4543bb973241b5f5c0691dc467ec0b63d3ce7c967b5102346aaa",517]
+["014ded5a50d73d54d217b3868706e51a125a01800debed54314ee9ecbfd389d4",960]
+["e7e13046827c44eabb3adb057fbd8c8e613df0eebd73dbf47fddfb670d16b558",2524]
+["93da2e37c5233be81fa8548db636a8f229c609ae5149b825cf5b6862a926e96f",1014]
+["c1e1326f342651543395b8e3a2ba39bca500340d57c61be3e5dadc592f16982f",894]
+["9caf11bcaaf490e827cfb1cb33253f900340dd18d6b9b40753a71f381a9b2186",960]
+["638c47e6d2acb2b7250ff8bd5bb4237405cc565eca88dacf514940393ed2ce2b",2524]
+["6353db46091f6d838d1d7f46f489d24a02a2b8ace653ea819dd31c0f4bf16ee3",1014]
+["0e985f422b155a2ee1caceaec83b75d847a80676fa400cb8b261c6869b092be8",894]
+["dcce1fbeef85bf51208c273e02f6928707bc7929902b56ebc6786e6628120020",960]
+["63466189517910cb0527466bf53bfc72898e878420d18d44ebd8b6d1632aa3b1",2524]
+["a48f38cca36306546031f219fd884d3111907cc8ef2914a29f339fa7b9020aa5",1014]
+["2a1c33900b72e45d1c58e5ecf8e4ed89131bbf0ee48d2b5778d557d89bc179b6",894]
+["1ad9e9874cf3f0739f03d0181d23bd3b2c85d9c1c2b79e53bf03fae014bdbf05",975]
+["b3db0aecfbbec5fe4f16704270998738c7e7dc75cda43364a601b7b22bb3a420",2516]
+["ca5f69feff814b03d70156fbaba4d338d5f21d86b4243cf702c305763bff3eb3",1017]
+["311fb18f7b8b71e6f98de09a04cb0ab7b8bebfca0b82f35375292321939648c7",887]
+["b1c751b1f4d5cfb310f7fb25eca150d3c9fdb551dbc7178940c89428f8d89421",893]
+["fe7a687050c45daa4c6f6384c575853b108bbc0b8eef6331367540065b60ad15",2400]
+["62408fd11f6ed8ba82cfac8edd4249dd77dd39298d9792b81ea4a3754720e936",921]
+["026bb9f2baa84464350fba3f14edb989406f1c224f066e8e21c9ac617a4a464c",729]
+["f96b98b29750ac8002599e40b1ceb4fc277ca1568568cfc8c2a53d3b5607bad4",923]
+["b01fa210ab7da84802c3445e2d565e02f20f93aceb9e9ab59d569788c1fe57a4",2440]
+["1963e2a817f709397b1545bb6cfdf3b20dd225143beebc4057e827d42c1316a3",971]
+["2df78443ab8d14f55f41b403f1d07fb4cfe142e0076ad4860c479b0f49f61036",831]
+["91bba2c2ced080a362c71e081910823b1490fd32cadee2e16c73ac8d18adbed9",2248]
+["88f74d0d7d7497453ed075fc92fa2e73a6d2ebe9dde51aaf5f5b7067a6186e3b",587]
+["f0a6f74b7dfc95012f2c0c3c11fc7900edf275059466f65b46a852d5b20c9737",638]
+["301a870c6d1df790a4f266e109320737c65653af14d99b7c77c5e3204f38106b",460]
+["9c8469571b697e7e8a73f5758300190b246877406c02ddb6c5a334a1ef56987c",2340]
+["5265049f4eef16b6a62d9f56d93be0876983af418cee523945414f1126327caa",593]
+["c039def11debaceb1ba4d962518ee2b2053fcfb5c36d2ffd70f986dd6fe61f30",650]
+["aeb322eaeb86a8d67abb783bb813e5b8e8146e2272480cfc755f4baea3267c71",472]
+["e076bd2e7318b2539c132b05f61a97c2e47d5fd870bd843cc4516f2c5f6f184e",2348]
+["9cdbfc783115246a2fa6368e6b9face9c99d0da31f0fa1ed014e662160cc2814",613]
+["3878e9ecc126a8f7e38710cf6c1ec4498f94b1ab83ea1c810d4349ba853b1b54",2360]
+["e6547e3ed16d6cfd289915b7e4b804177d93a05c9c331d9d2d4d87904a99624e",658]
+["06d9099a84233c14210abaf8dd23f09e7f8a39fb583f493faf607fe5b8c84cb8",482]
+["8cee73b0645867558bd2ad3212753aaaf0f479c7c76f9897a740710927a87931",611]
+["b70f4fc916a49e2675a00a010b3228fdafc35d77b891b7b74359f1db10c3276d",2368]
+["4e0002c6d2faa11dc479b068f4a14390de3acd238df79eff72d615175522b7f4",681]
+["3676642b565811f6af72fed100703ed82caf5e81d7b4a217562bafcbc655905c",481]
+["e9a1e63e940368ef84d70ff0ecd6c5948dc0fa00e2bc0887d6ed0990e2d9959e",611]
+["adca7fa4e8d3586066082498d1bd26effd37f7a26ceb0030bbcbf6d87b719de7",2368]
+["17943927233524410d96d81bd22dbe2cfcb996263e539ab191046a4dac4917cc",676]
+["86e63407f76171a124171d6f770275a49a8e7ddba463aa7ae8e72d9250f1dea4",481]
+["95bcbd823b79b936496e549d1bc27e0332bfe0aaae19a863308356537518ad68",611]
+["efd9a6a728ab04422d2095d44a1c0253a2c7b763a4de225aa7969d30e4305c54",2368]
+["3dfe69e18ffcd0c1b34a201e0265f1993872636771851fa97d232331f956734d",676]
+["97d8c03afbb2c419c595b97373d13ce38c01d230dc4be2e81a96f5415c87fecd",481]
+["e711a0d0a5a2701bc4e032353716c1bd80d2a01b09dcdf922a90c623cdd3eac2",610]
+["1729d6cbbc79172ba3389b8d451f73c2d586230eb2983fb092479973b76072aa",2356]
+["018ed19cebaa04a82aadaa4cc43c3d3e10c6f4553b84b4a173df37b30d97f1df",653]
+["b70b68ff882b017df725f48e29cdceeeca22e6bb7c0e0e44aebc6d54372dd451",477]
+["6a28f35b8f28e3c9e743c57e0e1d712a0380b1d01e0195b5623ac018410f5a6a",607]
+["e806da7d25314fc790fed1a2fd3e081ec7d73f7d2cc0ca4e09b35bf8dd8c9ed4",2352]
+["9c6d626648833d9d3d0965d6a22dbda3e75295fe6f8fd0795895a04919557dd0",666]
+["4bc023232da46041ce55f4c8a0dc2992841e44062cc0cb7f4c35fad34f54cdd3",462]
+["9e17455f7de130aceec7b34e677aae209d171421e56c3f65c239b20e975683ec",607]
+["1ab0c6fbd367333e216cbe48336f36b98baf23a80689be2889a2061044279092",2352]
+["6d3d2ef491bbecc92108d417c4f325ad98506d7eb06045d8db29fc099bb5f74a",660]
+["b93d67380e3a985ce845c522dc25633a3dbf7bc027b46afb2be36b3ec6846279",462]
+["f50e219a0d3e6f7d9582f05810a8ee3082008db1ee01234ef39436475934e1a6",607]
+["8b7f65d238f9cbbefc8269c2f50d717386772f7d2640218bede892c3887fed69",2352]
+["2fd7bb987bedc17e58977a4d0e675b2e5fd709b8107f0aedbae5d656a7c9f51a",660]
+["73ff6aa4d235eb0dccb07d3f8fbee377b64b203f63bf54ad4db94a5650ea36e0",462]
+["d5e2e6ac8c25ff7b3aa7507ec3838b6d937fa2b945e892ec43a60d5264822356",604]
+["4481cb08fcd007c77b12cd2320511bbceac8958b31032ce46b69160bb1f10029",2364]
+["2b894b9a2fbe270c30a10dc8a453c4e167c9d348c835e31fb808a5e719b0d6ba",673]
+["0488eab706dc053739adac2f90684ad76e18ae797fcfaa626769037784dfcd0e",474]
+["c4a6e43bb0ee82ed0f3fca334bcfafca17ac061d76c46fb903079748d949009f",604]
+["a5a429198d60cd84f6ce0d5fb24c4f7fda6f2a0c2147c2094c964dff52a98f18",2364]
+["64ae744d2179a3ec8f5d3f4f359be35a5f66d8c84208d01ed116e186921d420e",667]
+["8b57ce50efb84440903c8e6ed91769988e9cced3b7f9dbd96b13de53cbe73f30",474]
+["775a40eae7f6820970e34d505ed2a52453bb7466d18dba890c2290413a061cf3",604]
+["4d86781f614bcca225244cb7be7ac6fd0eb6038b42b4d5019b7e4bffeaacdc6f",2364]
+["8a804b9aa510a9cb2ccca10c14b7081d3c30c4f2b8d01a7a4f64823b335ab5a6",667]
+["162bb90f18484bcbd7aacbb388020639316136d1a3cb936d7b80ab9eea7dcd04",474]
+["b25511e8ec07222014e69e920ed8e665a8be1258aff5131723d79bdeebbae9b7",597]
+["669593a124bbc968ab1ab533674d6b67adfb01be5ee2466dfdd3cbcf0916ff3b",2348]
+["1fa7d0c97e6344613ffefd99dd11624fa3bad480935a0000ab55e0e53590c679",655]
+["620cda870fdee03c3a9ba0878ff65c0a91ed04e24e55fffe50c65ed4f2051912",457]
+["2981dc3b7d8fe0551ecbf4f65c012dae10f7aaffcdca42af729da1561f168408",597]
+["37e835f048195920e946c3bd65add7c26bd4b84fac209d3b4b25f9a2d1ceb335",2348]
+["b866a2ae9e4ab986d9732eae1c836bc8875f72e5c6a49ee4e73a1240a60d66ae",650]
+["bf2dbf6f1dcf0648b38d221f1cac0a950ca83b7cfb1e726bf2b9e498219a3966",457]
+["2b3875170c4ab8f77f9089f4a9841460d884b4cc433ccf5c0dfc00c3328a1690",597]
+["72eabf355a9416c1a0ec5b278ef2feafbf0e350d322e0cc5709689e6df3f03ed",2348]
+["5521d7b9ddedcc9486d12e6b433ff20387ba4bc777863fe013f6ae0fd8d5e43c",650]
+["cefa8dfad73235cf17fc400da552ca9e876a202039a159c3191f00603f0b1c5d",457]
+["80c8f666b19271b43ed7b9d9c7eca9bdf6df5c2dce8837b6edeec2d8468b2308",573]
+["4b68be4d6d57e33496ff270371d1d173de7ed152d499f81091886476830ad6cd",592]
+["3bf8a9954ba8d1c42b8a5a577b4617199da802850b3d1e7c3f0a0a0c4a606ee7",581]
+["bff9f8500f001027e13c6ad731656c5529f5779f0635cb30425420fc3db98110",617]
+["7772b925974f4ffb5c7c54626cd3b1d6f2d471bc53edd4bd2ce30d9d316bfb64",597]
+["517f32d4e2d81ae9f3acbcfee3dfc09cdbf9d59aa055192fe6ad69634ff04daa",2332]
+["55af1cfe36f5d5684f3e33203b921d50f0cbe6a2e5b5a7b47fae555fd3c2462f",525]
+["83e47a8a5fa91dfea569c1b4fb0fc2f10f0bc50b1e62a4a70f950cc937688ee3",423]
+["937d9a382cbed6384d8681f7d4c739db6422c484d1ca5bab477966f3f4d09cf5",759]
+["91d6c34401711a7b6580c1cf06165f9c28ce16c6b690eab4e012ef503370fbd6",2468]
+["20650230ea672662eb4cc314a8e62c2abdab6467ef9cc773d9043ceba6d0bb84",840]
+["b31ebbb8e33ea243bf541f8d9e9eaf466c5378405259db373c296c75d546bab0",826]
+["9f49927d82f4aeb6e387fbbe19d851b010dd170a6a71ccf4b346d5cb49ecb480",1720]
+["f6187252b3c9a9fe7b9e787ba9e4283acca4427a7ee1a7440b462f5566e4b4c0",796]
+["c2ea750b53e9476e118bfa978c5c5b31e6cecb4e4ca2a1f67339fa30ea52a728",2500]
+["6bb236a33d40d0ef8650227e17f101d7f4b95f6ff130691f8cac16cd12908d40",899]
+["3ae643137e8bd6c3073706492abbf8f16722d59b3e8a798f6b3b0d1f3c2d319a",1076]
+["c868b17001c46ab7d2c6c10e0738928e6be00dd12b9a1d6d692c1d10049cfd8e",790]
+["b762bc9d87d0e31573b63f6dba95e419087efedcb06aceb547e60d6f639cc208",2472]
+["a0050fe557173ea3b8153e9500d5dc54127fd936a41e50ce7904888fe6f3bafc",871]
+["64f79da5f55e68636848173f929d1441ea1b4c2463b0128a00352d3fd57fff8d",968]
+["5cd9d52918dd825a1f46bd6752b7efd0f0e23c2c574678c8a6a67e2ee6543a37",827]
+["25b16703c8ab79240993c9da2e4192d279ac6b68c8b33f7242fd603f7459081a",2508]
+["ed0c9ea609e88040b87218e4857d32df4f5822e0ef2287cda464ff8e055da8ab",930]
+["e626fc2319266c1364e94b40ebfab7331c2f89e3e3d2cfd14d5e5c5045ae1259",1240]
+["8eb5b86ed5d68ba317037a1e6163bad598033d40743203832b19c757ac0fd0a0",821]
+["612e44312cebf067fa72188c2a6ce8b19e893f38de41259ea037921c5881df8e",2484]
+["7a69c1e57886c0758457b433e681a55a2d50dfb20cb94a167a33e1fc0e951b89",902]
+["846f1cf0b99cd9b8de8268f8a9d4de69f03ae88052f3a0ee195bb97322922af6",1110]
+["5e6651e1bb6d2c2bb2d63b19a90b1ec0f520cd3d03425759c054f614d485b68b",858]
+["3e1e37536d0b35859ef44f509ed5e17658d81a87140af226083bf059a86bbf84",2520]
+["bcea385ad3b9211c45e3bc651ae030a7e430a31b40a1b60ed7f1d0460dfa21d5",961]
+["ad46b2c331c7a57e492e8eadddb6125f9ec248bea5270f7c0a33b0355e02dc22",1404]
+["88c0b0208c314374cd8282074ee21e3dda811b6e69514bcb3082aa96f30ce7b5",852]
+["53a022e66a5c96ad8a4c3d73e3d13e78830d562d19901d6e79d48adbf7218a3f",2496]
+["bfe7998250a719f5c9d272d2104385b4c54646483abfc2ca2787997ab3e851b7",933]
+["2003e70f6f41a96523cb3480944221813b8d60ac0265db16b81b01e8bd2cd989",1252]
+["b6d711066dacf255a64023c4fa941ea87a3876adf19e3563af00bcc126856c96",889]
+["7c2f76eb29cb4fe8c5744d0ae71fa81632b86877bd740634366db78b7d8dda0e",2528]
+["43fcdfa18beaacc6b2d195b7819d6984d25ebe2704627d536b52154f5568772d",992]
+["fbfca6c93fbdb2e083d2f2dba7c1d3986a2fbdb88e5c19e2792ec7957717c4bb",1568]
+["ecac4ffc90fd5ff056ac5c1b2c50f9020c93971f7f80f43cd48f2f40b91e0fcb",883]
+["55c9438f947efeb163ce4bb246b72b949eae0794bdfc665c937e7fed1d3cb242",2500]
+["687da5fc745815c6b5a09caa3aae77aafaf2885d7b124ad14ecfeac4ab5b819a",964]
+["f3373c232df0b358bdac18d45b52a87ae60dd41a7b66e13f4ebd3d2644f9cc57",1394]
+["ee978d494dcf9dff683ba49dd358f91dd8b5f3d97bd16ac76db06b93532451ea",920]
+["46b03a4148dd6fb921fc9ccece46b63b9f68ae567d8c8e65fa43ffd5cab712fe",2536]
+["1a1c84fce5bde19cd051057b1788a0da6384d319d1b05af074ce2627d3ac7875",1023]
+["b724cb70ab1934b89a73a0921036a3ee6afa5ed287ff42b049aad7e4dce3dbac",1732]
+["000a1a2d557c1e8f97f3aa6c668d7d9a2f21ffb3cf45335eb86f8ba51bdf9202",907]
+["e35e2b14295335c3ca35612b03ae28903a325893444dd8e0afd2e0ca4b8973c5",2528]
+["505e051e450159964ea13d212cec9148bf209f26e89d2068a997315e1d97edbe",1004]
+["736f18a18385eb3888a9b53a1750db122bec56724c64cbc3a1ffc8187aed7c32",1533]
+["d6b5d92b7bf7ad15efd1695c28ade0371f2d521faf3c154bccd0e000c082e869",2248]
+["aa9f0fa5dc9addd92fd1b2921590bae68d0ffd648d82f8ac6ef11b383b5f6f5d",944]
+["fd7f387fbcdb93697b51d22e61f77b458b1fd23925fe6bb80a77d1db4071b38a",2568]
+["d345c18d478bea14adb328382e2ea4a199a4ab6340ed0b53f4daf66c02874eda",1063]
+["0fdec8d69ff591413481c87b7f099822075ebed97c0462a154b1932f72ac6341",1887]
+["57fceb1462c6654f61978eac832a608a3e73ae5155f91ce9a9af9b0915265976",1033]
+["97633b719941b8d5c51719e1a1033982772b2f1e5c82c81c9c53ef9735bd73dc",2528]
+["e8273e6d8092cf2365f71322c9098012c5084818fc5b8cd24cee4cb21c7beb68",1138]
+["027d311431567da24606f4cb9e4d5c0ba5646dc001bcff1342063510ef8c01eb",1169]
+["6e2cb1899f0817adc9c625a24bc4d31588dd325a5bee25f16d6723bf00578aaa",860]
+["a27bda6607d8477e206699fe2b3690935db845f9f0600d06ce2471958722c9bf",2464]
+["368f1cf4d076feb0368b287e3802e9ee4d76503251ed2a291e8c0f78be912359",934]
+["7842a5f8472b9e860e53abf29434f8840c0aade2b0b9e154970f691f3d7dabe8",936]
+["222f48580303aff23f44f077938c3e41e93f014cd4b02b063262ab8ba34f0ef1",1064]
+["944a64c8df2c0308c831d7801f20714291dc440ae5583c026cc6c9fb14ab6adf",2532]
+["dda13f45befe70f6f5b83dfc2fb529a87853203bcef9a241cfea7daabe034a58",1169]
+["1ee478361c4cd987c578ce88ed35fbde2153c6e6d5d306b4d51756c545c0601e",1295]
+["23ef1d782bdb68478b163305fc89c41175c8bf64b943eb89a5386d4fd7f9caab",1096]
+["a0c5944f36b4f726c1b4b98a0343639563f0f125e28f85cd74fbc9a1bc3b8345",2564]
+["75729012a001c7ee6a16fc1ebc399d16137e477574e33c7c4ada01a0a2120fa8",1221]
+["06b960bec9bb0c48c4881504f63aaae5b5fa4a8bf0eae9825175b7f4d2ba2812",1583]
+["f76ee24d3e400cc43925c1bd3a3c0a24dc0e98c8d446620fb552b75b5c99c880",1100]
+["3a15f9cd7f24fff7631b7e0cca947dd0265acc1ea50b7add6778bc7e51d0f951",2576]
+["039b3a9603604608d04c90f49b67b9dc69e1ed54884495e4e81a3530fd3aad46",1233]
+["972794edbcfc9aac5be114e077efc992d31538cd6ad40a238c0b087b919b7a4e",1662]
+["734fb513eb2ea105187d3651d70484f90ba03b9796033e82af18837b74322571",891]
+["a9022f808194083687c68f634b8dfeb802e130f5954c897895591470adc43acc",2472]
+["7dfcc90a53fe5c642880882f46ae2b7a07574c01ba0f0c1d05c8f3f240dc5cf2",965]
+["40b114fee83111a94d49f98ba2ce7a5cd9f3f1e227f5bf4df1487b32d0a8cba1",1062]
+["080e82b04242bd4abbe14477f44a3a775bfc698e878668d9f41c5ace15126c96",1095]
+["2efdfdeeeb8a44b0bec68a43ed41d9b8a50e2f76bcc8a91e9ef9379b65612dfb",2544]
+["bcd725dc6a0498267a0f50d039e35486ea160b6e1c41f80709098e177008b224",1200]
+["b63c24239c6a26e868d3f797735cf43293aa72fe36bb4d703d97c945c07f9d57",1421]
+["415222b6ea7ecaca5c89ccb83a7f01bba0d9d0ae449512a718dc1f7dcef6c3d5",922]
+["9fc4ade12fc4a5e05ed26fd303247639ced9e8bc8986cd61f7f6db77eb4b0fbe",2484]
+["e1baca04ed3d6bf9a14db71dd064f14f3973b3c681b5a6d0d46831d4fc18aeb4",996]
+["a3663c3e3b855ae7432b4dcd9610af02cae6a5101badf57c11276dcd7f332fce",1188]
+["de917690753e7628fe1eca5c91110b6ca0d03784aa01e0737d892ea8dc82cc9d",1126]
+["89c7cbf843a88fe01b04bff482d60fc0fba8af967d3be5cd21c5d699ab3789ff",2556]
+["70c46131ff58d16add50060ac8e33381ac904891459353f193a8aa3b053fd838",1231]
+["3eeda494f89d31b63c24dfb6659897b7b586e456ffad20e7e7c0705c45a41de1",1547]
+["c7df477fb869d89c2ca2f9278a389bf3d354282ff96065c93e2f397bc18ea7e2",953]
+["5d1c08e8ee4dc9901a474890b3eb3b005e69647b1ae228a1a6109ba355b042e5",2492]
+["68798080b6bbb08da3bfabae29894ef966ac209ca66ac1e30fa842c33f34fa1f",1027]
+["e49772e1d11e03525f5ca09719183cf8e1cf5d2cb7ff8f772f7cc7799ede351c",1314]
+["8b90dd96e8521526066859985bd1a0ca7d0322631cfb27d40f6fa113c7a28ec6",1157]
+["b09b32a2e28f72c84dfc63abf7a30693db3881a6075fdad46c59d92d79eaff47",2560]
+["a0da5e74f52505425a8dc9ce23921cd6b403bc7249a664c4ff88c1355b67dee4",1262]
+["42386c4a1dde4f98202d61e35b2128cd4b61a73a6520a1485500ba2789272c7e",1673]
+["7e34c25faf8c0173993920948c7241e50e78a066cdaa70374eba858ec3daed8c",984]
+["087fa5ddc9cb081c51d1ed4ace2f9e283df6a7267b82dac432b791f47aad4416",2496]
+["70345e747a8e3a77954906bd7d903dbb14bc53a01542e448d20950651a4b2f52",1058]
+["dce9f01e3a0a3e6655f11608b47ba28ce62e803eb96cb42c7c6bd5fa69a8b92e",1440]
+["a2306a263e415e7b1a1f685e64c02de8202536d5e1931d4709b4316875353fae",954]
+["8266bad33bc5981006452fcb7d4c812158f0d734a579ad0bb4578c286e6d5989",2532]
+["a3d8bcdf2c0c38553a4dd0565af6dfdb788d5577d9c7f215fc2be88f0d0a08f3",1056]
+["71988c9586f2cc9087ec2d498c86fefa1c30d1f906afb4bcb7f573a5fb91fe1a",1386]
+["16b290d5ee7e56a5b86d1d3c526be33e794433523e937c7241dda9516b0de524",985]
+["6ccfc633b721df1ac964976a32a717a416489d6686c790db9b31fe354804d8d5",2540]
+["2867482dcfe21bf176d288d4874d1213967fe82d11099a6e98d5cd47c52de23e",1087]
+["16426442322bf7811b6175e080d86f086d23768773740b516a7ba970b9100b6c",1540]
+["554cdabf638932392f9921e4c64b57a519d0576683ca931d1893b5e394eef8c6",1016]
+["dc47933f4da9e1bb2f1f7d4d3d8beb0f21c8b079edb9f692ca06f9562997b4ab",2544]
+["3da09cc9f93f6bc987892e07ff2d7b7dcb2a666e8d937641b28179a94d80c11a",1118]
+["38cf600d0eeb3e51ce45c8fdf3c8f7caceafca5b9c7365ad35b63b190f071060",1694]
+["f3d485b1ec5160dcee3b61b782754bd421c88f292e57f359cc85bd8ab823a4a8",886]
+["900d845050cb6c31ac715c02c0b34eefba910ad6687c5d89ff79373db8693899",2500]
+["09c75d2e834690d0126c73fb58b2d53e62e30abdc02697ea7979b26d878dd264",984]
+["f9540df16497f04d43f4b4ef2ded8bf393ea18a279ad55fbb054ac0537508f00",1048]
+["90a18cd325a0cbb1f45b037e435d430d886c194f27fd4c0d2dae2ae31f328da9",917]
+["89245810216b36759fe2f60cf649136dbd2e6123bfe5052ee2ade6d85a968969",2508]
+["542188590bab58bae94b3fa677b57aaebc9c5e25b9a53d22a09e0c11c993c71e",1015]
+["50a383d294d02ca693ec94dd874462592290a890cfeabfdc1ac860591d782041",1198]
+["b1569086e404aa4f94f3294d69645dbf9345820c62b1ce0d46c677ed3259d4c2",948]
+["77889b550a5c49decb256c897d4c4869e3ffb2d0a402127767a83761b25be262",2520]
+["50adc4443f249da84fbd7a497dfce9e90d27d7dab0916ee2a2edb63b1979c945",1046]
+["60b56bfaef9f6160992bb68e1ffdae1108bd2222fcd84ffbd44d2556e25e56b5",1348]
+["72f1df86a120801c033d55867650fbfcf1e9bc2a6da8e2a0e979c4b4180e2189",979]
+["389879fb785c51399fa339872d148e50a62e728d00a69019b59af4aec686e317",2528]
+["21e4567b6ab0d1e297dad55932f87f829454f6192f595a8d6264fa120f2e0446",1077]
+["c9c8d180a859fe3345fbedcc3855824bf0caa830612bc6c87778967dee0311e2",1498]
+["7774ebc2edda17f4030f31c387ca3700593a60bc5b52da73980d299ef922b064",1010]
+["e78f5245b8a432a67d808d2d30e1647cd35e9327746d96b0e99bbfa76f28d6e4",2532]
+["9486c74e945c899c9e565c23ab05b56650caddc0a5442e24562f0edb46ef6f62",1108]
+["9444cd47231f92f7f18e5bac2d853240e6b53b9deeb090e891ff2eb9e70094d6",1648]
+["378219245b0ec4724028c78618781af1c46b667eb754023d6ba56083bba3b89a",954]
+["806e1b1c937ad5146a532fe98c1b01a867ed083f25e10071aeb84fdc8bbeac55",2532]
+["1e969ecd518c03530e217fadfd7a68c1bb5410e82f556c46d9838f2ea67c3c85",1056]
+["654fcc28830d9a7d1bee0dc0c010af03aca746b77a3f8486f245aed4230ba261",1386]
+["d69335f4389f203557519f1ed443109892915851655b53f2f769659de2baead8",985]
+["6f1788e538a912240c9482adac4c59930baa671caac9e829a521941d039d7548",2540]
+["ea6d4123d596ba1a6b17ad9c5ccdeb3bde905924c70a4f400f763d02117cfecd",1087]
+["2d786feb65ae92e217598d068d510c6130b0af2a9badc593d60ba365a90a53ea",1540]
+["1341ccb8577d00c211dcffe990b9d1be2e3c7e8061ca04796240313e5b842884",1016]
+["92daec2749ceeffb6ecd19b61a445878d5570dd9dcc0d7786702e9e237b8c04c",2544]
+["f55ca7d22bae9daf8a9da4ba4d2a4069032b8e693eb094ff3644d85dda57feb1",1118]
+["5257b3368b14093c8ebac40d81e0e15e1ca58c9616a2030f35e1ae925cfd3c3d",1694]
+["005c6ed3121b17628e4ab2e2c9f9526430ea43d2428f1c2498bb15fdf9cecef0",886]
+["082893e630dee4fe6e427edc18343fc829688720e59b0803f0d15aeb2e9bffbc",2500]
+["676dc8a178fc23053ddbb584c5664a41a867e90b1d3a4450c5e9350a0c53c922",984]
+["84350c1d9152485cbf48060c143ffacfa9505093c2efd6854474d9c447c36433",1048]
+["bbbc2b0df279f4dc3adc695b2a41cc7fb7eaede0b1a74b36e831f0421721f52f",917]
+["049339a3bb9c1523b2dc9101dcfbae9965ef978e240e04b8abb0a61b9acf3c3e",2508]
+["6efd640301f4491da65b6f7e440ead674bf7f0c1c60119a1cc37200a833a3ebf",1015]
+["dcb047b418007515745388fc75d47550dc59109874fa8705166d229c8bb301c2",1198]
+["f171a1a04ee44743ccbf5922bf65cde781305391a36f4cef1bafe47e9da2fd58",948]
+["0418a853ae39da141b455a44846445c8af0cef28829f75deeaa4a01b7e67e987",2520]
+["845d0b2516f48c820f43dd51ac1239693ecca54c89e3b13af10fcbcf315c5982",1046]
+["3fdc6a6011b389473ccdd9224e78700902d202cd79aec47dda878516a7fc04e5",1348]
+["dbfcb901154600a723cf62e2b1e11372819a6a3eaa7fa3bfb7ef5e0dd25a2b61",979]
+["2447662e354f0d54ec709c1c235887ee61cda97067490ea8c50e2a3eb457e57f",2528]
+["dcb7bdb8c52e1cacddca83e5c7957fa607a17de139f63b18328aa3d4520309e0",1077]
+["f97b5d9367da79ffd7201086a9136f81c82e75421c58120aed59137d1e4c01e5",1498]
+["da91b6a34d3516c13c3c56971c4afa22be38dd31157d1cfbf54f655a2ee1a904",1010]
+["5b8c7d55bd70ac89f7ba9876e5e2856e6e560f7f046acfc902540fdbb1c44c2a",2532]
+["0fb647df211093b31619a4130da3cfb8dabb00eb013275235e9026c5e6b90f23",1108]
+["7b5f70c430abf9482bf6ed1fb3162856a0c213cf9face030b32c2ec641ff8479",1648]
+["f2dbb4d55e1b38049c17e2ee27b6a713c5144a97122e6c4c222f0b89faade080",954]
+["c71bad93db42b02c9d7eecb485824334fab0efe78e06214652dc2be77aa6a2a7",2532]
+["0330b29c39922d129325f34ee92c4fa0cd86738f08445e066e9b290b27b98e35",1056]
+["3712131077b74832494e1914fb21983456039a7a1ce7a7f31954bab7e2fb6b6a",1386]
+["17749c43dbf22773588de480c3feaa46476252046939296677e9273566b72f03",985]
+["dbf2fb1f5332a8ab683304cf178e3acdc46fb0f7857f0496a1da92f66865d89e",2540]
+["c9f8dd1cb4582abb040ad41d0dc2008aa586594464ee18d920031c9576b076f1",1087]
+["7747357ae471aadb621fa97ba75050aa2a3e88cb7810bebbd888b111d4653e88",1540]
+["77cc96db058ff4838ed403387858e9edef624004373d54c6f2d0b2880a14a40f",1016]
+["7a2330886232936fb7c4cbe9114fbfe7d3d3a215e2f14f49d462e212ca625456",2544]
+["5a2e83aeda422e26faadba00e73ce3baa98f784b9964beeeef4a5e31f3771d7b",1118]
+["e8c35952679fbae8866167bfd9525ae5c2ecf78d54887ece15e4e3d1634e3b4f",1694]
+["25470e77ac60d3d4e1aa71470a968c47c4803e38dea241066446a8e29c7dd5b5",886]
+["07c1c6985f03cbc591a65360143c96d0e067e078e102d7bc491f9c4729e28ca8",2500]
+["1b8eb7da91716ada53d72ae9d1d51d971d408d831da6605537bb97818c152a04",984]
+["3dda5432eaee99bfc40045660eee3666e6298049b0a72c672a8a80b2667c360e",1048]
+["aed17adbd02c07b8bb1b739e5c20904cc26dc1c140cfd852542017e2840e7230",917]
+["ffa66e4eb117345e5a3957f295c912de7b1c268e62e54d5f926a7deb0f15c7d0",2508]
+["2a7d73bbc16cc317b5210ca8cee068356708880ebaf4c5abe97bc26a1e541206",1015]
+["c8a8bf78b1e66ccd8ff61379d57424c619939e83d689aab3625f98bb51b54b20",1198]
+["59d1fd6d6e94c843b37e39f2b8fc3bcca9d62e95e8a556a91046be72bb697fe1",948]
+["48ee470acb7549775925093ac0757f0586fbb16c4003b00ff228fd4fa5c20bb8",2520]
+["19ce645dfcbcf3290ffc070bf671f38688b8cf2529a66c7fb84aca0709444c9b",1046]
+["d20fc880c5b601b2fcd855afa504052ba6dbac5d29713f6c7a1058ba1d52c2ac",1348]
+["8e912fbe139edc3d0f6aa5e82692b816cc918fefd27908230a76c90386fada6b",979]
+["f88e7a6d5800fd14784f0914d3174d7c0742c31b7003244ffabc1b9935ecc6c8",2528]
+["27695d60b2dfb18f8d61ee446d984c42cac5bd1c23be7a32e1790bc139ad0d36",1077]
+["6e285a38052824a09bf25a881771094ad52cac82148d6497e6d14fd9a10b1cd5",1498]
+["3eb15ca14c65b9d84100d785f870c440e6ff44afba874ab498da597f9438f4c4",1010]
+["dde4deb8c3674f3f9deec4e6994355640ef15decd4d5e3eb7685c6ffc56c2c37",2532]
+["7f859aec3ac06032842d8601682298acdd1592f53b1e5015738aa39e4fea5763",1108]
+["d8ce8eed72d89af81e4ab91f3ee8e0412912c06af4d28ab7ff322dfd784f4617",1648]
+["473a16dd243ef35c359c2a4f299e139a6ca33d7212d31b3d06b9931ab568dbd4",954]
+["9ab5ab287eee38f4285a0d7a87ce5b1cccb8757a61a1b9cc262f8fe826b23461",2532]
+["01f6b905ba47134213a465ccf4787fe54da49793ea7a62fbfc80180640739a7d",1056]
+["401f1824e1b2bb738275f709a8bb84c48ca62533f1246cc5ed02c1c24d3fa189",1386]
+["ff1f4f5bbb08c7a633d6c157917dc55533f1be1aaf34b17a89c8a104d7913b78",985]
+["75d46c9c6c4ecb8d3270e1704bcc8add5744562ba45b84db3aaf415b5a52a602",2540]
+["4ef17f74bbd6dc9b25a25dd4d2eac72eab22e0a4f520f9662ffb32a4858998ae",1087]
+["06af6c4b40c9d7701b034f0846e50a3bd98af305232f00a53060f06549a8df02",1540]
+["43179dceb91be75f6e9934fd208f6ab896275c30923df6b88a94231b09e6a3d2",1016]
+["cc931d24bff91a384131f74a2f4de51b809eaa4aeabfc0b6b28b57cbc1adbff4",2544]
+["d4525a6b0413fa2c64310b48dd6de89c06f692e347f87fc99520366765643477",1118]
+["07ce1f687b4be0f4f4092ff3b39375a69c730e7ea431c03d8bdf2c5d71d40ae6",1694]
+["081d27e936b26e7d927eb49030982b5cb80e0a237bf1ec946a65c05082a5660f",886]
+["f60d927b64483bce2c89bb1bade397cda8b8aef2effef49f4d6ceb1e3e201d33",2500]
+["7a0523da938d855a95b83ddb64a339f88ced26473dd87e39e34d05c8ccad9ff4",984]
+["d9e34fc645166fb2873c8dc01dcda484cf7397e5ee4fe46ab60bbdf5acf1baa7",1048]
+["6dfff98b85f2d9b5a7d1abf1908c8bf40030ad579990040afffadea13dd548ca",917]
+["3726623b1a5ac936d76c3c26cb1b4817df196e1971ef175e9c85deb308ca601c",2508]
+["c67a5b4e8b0510cb83bb22248fce070d144127514e1f759b44db98b43c5c4b8f",1015]
+["57e93548f32c9d2566fc10252ab6a606dd38e76232f247235dab786be83c91be",1198]
+["175fb3fc52297db0f64877fa9173cf6e725f1a67d793b6398cfddfe112c5ac8a",948]
+["9f21856bc210a6f2fc7ca7a62e7e90ba465a985f81c260a99322c8e216ce37b1",2520]
+["4f7ea3e5d58d0ff5c5a1aa6aca1290586ab9010631d2d2b2f75a158b1a1b0135",1046]
+["ba4db9d6390d57de09cdabbf970466b6bbe4ecc25a5803a9a52e9e9bc8b00e3e",1348]
+["30f022f62da6893c3cf16a222c20b1d6493ff2dd2e3725237fca187e91b3bde8",979]
+["5370e28816370f9049c28164c8800ac7e5e4c2985e8df22710cc27ba6d993715",2528]
+["037689c4ad805a17da30f7be07c263ebe6b139389f06587906e95805f8f58d29",1077]
+["b23f7b7acea739d7d9aba83d21bacc903065682fb66b6b6e07896aca453c5bc3",1498]
+["b94fd1f28655b7fe5a8663d030d965da59c8968e90beef4e488cb838db8c26c4",1010]
+["43f6c4bfb9d2f5f54c8b934eaade15330cc5034af00d6f5c22c58d635df02ecb",2532]
+["530f0b8a8003a0835994b9df08f11338e3189a8386f20130d934530af20501bf",1108]
+["d61e8aee878a5b52281fdb58c37b22b0c9ebf06a9aa77f0eb291607592cb8c00",1648]
+["6b33c2deaeed8d7e123a17a08d8cd7b18d5678c714e00c5f84cea60c654b411c",954]
+["d3c4cf74186083b9cb4480cb126b7090bf8b13461d294ae34672611befadc40d",2532]
+["68784daf878c726d162ddd6d0ca36276da2ea12fb6636724338d1bb8a0b52cef",1056]
+["d4d7557622b04ebb00553ceb9da76f8c5e04bda0f3fe9883beba0f02edc82bd9",1386]
+["7779af4c4c88f2f3bc70a6fc8b5633d5b7655490c28ea9c49339cc6fcf02b420",985]
+["5ce8697870c7a6482044f3410e368f287dc1c219928208101e5a882a6cbaf361",2540]
+["de2c32ce091991bed4c929386f64d25fbbaacd634d5de3f06e566bf039dfceb2",1087]
+["7e41e4c28d43fd3c5b19ad383cf37b4498bcbc422db668fb74ee50a361b26dad",1540]
+["956e59811c6fdc08a6d1754a219620c549356b0bf18376f6a5ae7fa742e363cb",1016]
+["fd32cbc4aff1acee0934f7455d7164a5d1dc2738216f28608eb0dcff90b1194c",2544]
+["dd10522dcadfccf313444d33690d12a2bcd3d649918b4235445deff0d5f31d0b",1118]
+["a7cd36c49c90a2677f3ca51ad9b9b736de36838ed3a30e1145ba1d453e63e27d",1694]
+["7af25df9eba669cfbc120fcb028c9d8d34ccaa01481ee773342cfe1b6fe7bb9f",886]
+["d744176174bc4d80890aace53cd4ff250161f2b4a7adb8211338e89be62bf503",2500]
+["c3d2fe4272c3426cc6c96c15223c1440b2ad426050d9f155d7e8f7c3871a1905",984]
+["b22bb16df25a8e1b2481b5c263a3b05dc0d0a8c6973384989f40a4f5f0494887",1048]
+["ab5974302357d4a53708761e1a7857c5a754d0f658b829935734033ff5c1a784",917]
+["73447fd7344490473def3f981949a33ea869c93189b253e9ac44c3c4e2e2daf0",2508]
+["8782765b8ae153cca922c60a45c0a02f742d09216272afcf9b1e8b90ddfafc23",1015]
+["082c4304b77d40d3f33ac0016f5d2ca8b0dee200fb6b0b34b6578e407c90cc64",1198]
+["d5ed1fb96a57194fe141d78428a80287f2a49b7d7d078457086932f95f3bb62a",948]
+["2ac5f0303e3bee979d6df7f34a39bee2993821238a2dee16884bc6dd2bcfa213",2520]
+["2b05de1e7299bee12c4624e9f65d9e86cfb61fe99e0e6009bb0e9e5c5ea8f360",1046]
+["54702dec334671640b2ac00f53ccdfb7617f680b0147ee7d8892f8235256dd21",1348]
+["ed5438707870c31851f9304921afd70d2b81b06c3148ec42213f609acc60f11e",979]
+["d3af1a382ef08d7ebf94b3582340bb63a94e5dc17d8eb8155bb510355f3163fb",2528]
+["390b64fbb37cfdc344fb6c344f136819ddda08e27a1f2966d86e99c6a88894a8",1077]
+["e10131ffc5cb879e46fb12fc69728debf03d28b23d292dfa0bf1b9a55067a401",1498]
+["b9b127e2746ae224c5bf75659bbde8143742cb9f8276f24320f853e6a764268d",1010]
+["a76db74e8297ccd8a32d37acc975459277e4bb53c5ecc2705d484ca546af91aa",2532]
+["85e3009b3cc94c0be15799e36f6cd2f6618e992c4adfefccdfb5884e566fc6c1",1108]
+["4d79f6e1854426ef8b8f8693d1e842172a475c3706eb259393c00177dcf647af",1648]
+["b2d154bb19b8a99743f7df0d5b718d8eeab1f29cdab6a36af9e5679490e4913d",954]
+["6a4fce8ea6c2b9c007adbab171baf3b2862967ce1f9dd5815cae2111bd15a108",2532]
+["81124b3cb4da6881736ea90ba0fabe4803654ad1e8b21460c4e5b7b4a5e7ada5",1056]
+["8fbe847d18552d302f8ea25a368655ecb81eaee2ba2c72b4e158191b422673e4",1386]
+["9fb179bfd58674e35bbc6ffc64c4203ff3f383c59ce1d08dac4c0d72566a4ed1",985]
+["8ac7229d946f8a0fa3dc145e46712e586d7d43994da6735ede06039570096248",2540]
+["82f639f22d58620b1d830e64461210d65dc011ed17024ba6b078151af8ae2b04",1087]
+["d410c47057cb1d093deaa68e8fa9d0cc4827831bc54989a59927527bf3663c6a",1540]
+["c7bcf90615c5040b6ea5cf58df17b2f8bf4e8f0688d8ee242f68cc3b2366df58",1016]
+["cbecdd6a1abca678a6f9d1fddb467d74b8caec6d300160bb26584fe5ca5489ce",2544]
+["87e31842dca1a14c037ac81ada6d6e008ccc1c6d6890dd223972179a975488b9",1118]
+["b44f0295a0ddd9a65edeca5028d459e428d4087f336798bbd05985f55f60ad30",1694]
+["df3ecf84072002a0b5b8c394b64829e0664601a9104484bdd2bcaf980d03e88f",886]
+["9601a7b55bcfc15c15d0669d9d7d5e17865e2b1bc655664682b7541172fdd117",2500]
+["27540ce82df012a296ad7c1a5113a213c80ebfa6cfb250303c097bc30dcddd1a",984]
+["9700c1d3103ab1875c56954ef562144bb0f322501d53e441c016dbb1611b07a9",1048]
+["15fca5f5069641af88b6ec2f095127b6529ba2e21e3d60f50b6946f48477c169",917]
+["3a459c4131b8fdc3ad8d92d1f655e699014d8b8afe325f787a97d2c13f214b2f",2508]
+["167b893076f51f2545862c1ffb8c66e78db7ee5a9bf26e455a10246cc3e7b0ec",1015]
+["adbbbcf16a5036947903c750877cfdad82333421583ea45bb0fa9129b3464d8c",1198]
+["571c2d19a335d87df7a637d668968f6cede3d683f6365df2fde91cd3832754bc",948]
+["29aab3a787a8ffc64981046f7ecee2d5d43164acda6b0937169084deaab587bc",2520]
+["dc3f6cd8bca60d1173a90bedfff362dd44d02b403535b1c954c09a6fd96935c4",1046]
+["332069e3848ebfb20b437322103bc82ab26f7a7e956afde3f617b4cc73ce4369",1348]
+["b503677c269af3745bc4dc167001bd3f75faab22af1677268b52f57d66819733",979]
+["e065e6c717327e6ded8453fe981b3b0e5f13075aa5174360654a9fcd3314c6aa",2528]
+["602c537cafc78510739a686ca10326ca53f092ecfada1de067747fac6efac253",1077]
+["da154aed3071e5b488769dda8fd1118b6cd596c7a3b5957772d60461876f0b1a",1498]
+["196a9c8312c558e2fde3b369f1016e6d0e3d13d5ac737857836a25b160e10241",1010]
+["4e32811ad0ce1bd2b7ed634871002d068e89b8b0d08856b6c46f2cc81072afce",2532]
+["498678dbb223e03025f7bc86ed5a66faf04e5cedac1fe17833a5b6e21ea900fe",1108]
+["10505eb841bcb013ad7ca55c5449dfc4cffd80e7795b8a10565bd345d751e20b",1648]
+["7a9f5b7b9535044cce019f54edc1a902035e3828fef45962d0fe7d5652269cb3",951]
+["5ff497437ceb9a1adc509f55f02d78f89b46a46496f91af9ad890b26695b49e5",2528]
+["6168920efd2c756054008cecf4deca6eb02b9d2716e8e8e4c7494ee69dfbeffa",1051]
+["83d583cbdb88638a24329ca79685b505d916778197e07965b2c14fc6e171ee30",1371]
+["1f4ba17c6be788add26fafcaef7100dfb667432268dd89843c3d5eab874dbf8d",982]
+["4f4e81436b7292fa814804dfff46d514791955ea98ed7516e41e182e11be31d6",2536]
+["f448d297eaa081e9617468488d9d10d299e2d4ba44bc3ab9af596d9ec280e0c9",1082]
+["058c348bd1020b6d8e8867505c6cdfd4aa1c63d00e0574d8e349349309be22ab",1523]
+["d3d2452cb63bf32924277e74c64b74e3dcf34f067cbf4f05624b561d24914114",1013]
+["2d46f10d6e29240ef52d1247ca3743129b6b0af6f215e4ae5f197b670f2e2732",2540]
+["15575928c2217cae0b52b184206b9573b426d4afa8ed17c38e5e7c437d417153",1113]
+["9bad6f3166cf8ded23e431497cc2ad1f3366bb26eef5336dbce4111cf8ea5ad9",1675]
+["e2f1abb2bae32da32749be0f5314204514a84af2b5626f49b69a3f654fd76d20",883]
+["e38fe4e96fb844b638b11ab9d42a60571a7ca4d356e08f8c0f28e88c701e351b",2500]
+["19ec1bcb2941cbd92c5aa1476cedaa5667c43ba21448e08b945d8eaf0d4d8e70",979]
+["3224b1aa3f431669a073b3b4f22bd1c22f6baa1eee7973065b763da050eca25f",1037]
+["3c5d1bab4193b6c588db063638014bbc01b0c2d0a31c8c61aa6e2eb1af0dd716",914]
+["1a1b06aebe25230ae460090ed196f4270591a5a61ca8cc7971b36dea6662b4a1",2508]
+["70aef9863243575c0885f5ed23c64be3a2163c9564cdfa7e30ab0773acaf98d9",1010]
+["3d9deeb34f3b027b6485706d49ec9eb6c96c239c2a8fb5c13fdeff99699724f6",1185]
+["9eabb0a8fadd3c36e813c517fa17c8500510da795935e3908edee7e7d5ce4c54",945]
+["2521f6c54f4e20ab99e94c18def80394823fde9c774f3cc4cdf3511eb50218c5",2520]
+["80a29261ffd0681354c2c53518404b2af9bc272a77e6b7486df279c3cc855265",1041]
+["c1b3637e50514926a0c999124ff79416a8e7cf992e89864ff16be239385a7a2f",1333]
+["86acce302dce2b7c99b120063758a7cb3fb8bbe0eaaf328506eabffeb1f7c178",976]
+["c153f6e28286bd180dc215de07eca7b048da9839f92819bd961d7eb01f05b8d5",2528]
+["09a9c8ad81f757cfa4dda72c01c99dcce108c32fe271dfd4fa77258f10995b31",1072]
+["1dbc678e2ac07a9350f5872d7ee72b747d8f8365903a875b0b92ad212b2cf480",1481]
+["a0f57edf02a429f2365620c40fc5f04fb189ac5d8a23d63d4b37c897f11e5ba9",1007]
+["72969d523ad74bf536cd7f87ed56ab1b9fa479f711296005792087bfb88b3e50",2532]
+["800d6a60396309d98a2a029b66af40cf739da3a9a7c29698c435786208560e09",1103]
+["87cc54a4242a52d5c51c36e68136b5b1338ad2e3b9dd4ea3ab935689adabb72c",1629]
+["1b311447afe627cd5f32661bbdc9e900d21327e4b05f67dad0a73da38fef8da9",954]
+["ef5f9d91644cc9485c7e5754510a4774831c540a2431b7f0b35dea461339a0ee",2532]
+["aa1d34ee9bdbecf70e6abf2cc4ac1d9a01c8d3233e70eb2d1b2a432ac726c87a",1056]
+["da4a716e71a489ac1f2c65052cff8013651d29b3b7b620a7b19b82bb9cf380a9",1386]
+["a59723a7e76bf8bbd75676a60e765c92968dd00812d0a490ddcd93f5942fd4ae",985]
+["300c81d8a80e1df4214758c7835db2ed9ccd2f3e54f0d653ad3ff5ee46e05154",2540]
+["b1796d6902f7c03729d25a0f57e7837449c5ce1baa56f017fdb4afceb8bffd54",1087]
+["b6465fcef3c78c76ec31526386c4b7e945d669986f54d3cdb189302ed60a526f",1540]
+["07162aae857efe0b644dd079fa9ace455e2c4b614a610d17a74938982e17c640",1016]
+["4fd5bd3cd1b938214d4faf534366f5d7c87cdf125c0ac486b8c74026c48f9be0",2544]
+["5079689391d8052ba99a1c7f52297f0cc71674b79d6b51297f690057a7e01e64",1118]
+["5ae33ff42cb7ce90eead5720450686dc30561abcf7cc527b85e7f31234a1e99b",1694]
+["027ac59eb90f6dbe514718da10e6aef4d790325f570c2b8364593ff2cfe5b98b",886]
+["5b90ebae259c6c6d58116f2e19f91888821e4c3a20565c98aacdd2e887882d32",2500]
+["f8bc94180ad02e54dc18a74d87bdbda5ab8648ed634d46da93dda3ed5ecbd100",984]
+["536cd9126d15bdd6f6cf5db7c3fb6e5ce82ebd16586c07755959213e09434baf",1048]
+["1da147d5113fc2f75c5fb8c9b151564a23196d5594f50211c964bf42547e4bdb",917]
+["bc97bdc95b3d697d3c53a99c8178c7801d396a39266559da549640b61299ff4a",2508]
+["5e5ea10ee742054bf2229e8ae6c32f8491fda019946846b7437206b2c6c12a63",1015]
+["7fa78d2132c4a64de3e6a59a26c0496e592e624a62739bbe1ec90091c2d68145",1198]
+["c79ef430a4007f1dcc659fe910ba5303a2207d0eccb15abbba38343a94089984",948]
+["062e4daeaef7f4298a90573981aca5104056d9c44a2a1f2ad840d6dee7452d96",2520]
+["bcd1a6f5e0d75a66af68f906ab08351833d499e027dccb53560403da2c9f3690",1046]
+["8fab7ceaf50d9a95b648cab378880426776b305ef3243f21d0730dd85c63e405",1348]
+["4d2a4cafff15733534bee3f729e79ba5d7056eff8b09a73ae52c5b60fe86d206",979]
+["71d17c37f70801170838a167761f75279d692c70833728be39417eef97db2c4c",2528]
+["75fa08573dd25ea231bbc3922ba15facf3043d5e916048af575a9fabd97586a9",1077]
+["2ab2d0e262c5e672e2ac974c791ea619407037e30c57349c4c9bb6b843ad5580",1498]
+["aeec07a124f84065c521fbbe61e03941ea1effa40f84e870af25901a2e543775",1010]
+["4f22a47101006447bb902031f4da2de746c67edc3c8ad1b40e23cc0f45efc2ff",2532]
+["1d02bc7c40c954bdc47093694cbed7590b12d738bbd5fd364610340902e953d1",1108]
+["7a3fb324f70d71f854e09593b20db8f35312a97c30d5383f70c1ab38be4ed930",1648]
+["b42f344ec70ebca8ba214eae674d810083e460f4e7a146f3bd553b22e809f377",599]
+["9aa002701ae707e01b56fa07afb17c8227cc1691048a304ef7bba907df7bf07a",2344]
+["48a8fcbd21c183452bd35af659981a459176314cc40781a1311240cd5bb73436",530]
+["c1a954bbd1aab06d0c13275ce6c5113bebb1a7ded574c9957c36799028490032",416]
+["9356260d67e7c08afe1199fde961d2a7d858c961084b4255b4290784a21ede35",614]
+["51620961e157488b83e2d321113496985e062e5adf6ad2d4b49bcfa13c33fa37",2364]
+["b1fb1bae7b6c0aae829f53ee3ddad287d29fe95afb80d4ef276804cd9b545796",555]
+["357a47b40909e946bb9c023e7363a0be7ae5b1ed972796a93332b18089709544",441]
+["250643afcea52038ba69b583b0ccab165b16588b3824f5116da4b5284db98115",845]
+["0e612a3e38e8e2c71210df1e4cda7993e073b99c4983ba595568b8ecabd686f3",2392]
+["7f484106e9be8b282934cc091616d1e2128fd34f28dd39009c46c30270ad14f6",858]
+["ca6e212aff73e24e289d4e7bc7054b926a785190cadfbc762eb72489d4edb212",673]
+["a39f2422339b1613ba27aa8a9b12c69c93911ea9da74a67f6e3143e245938cc3",842]
+["1b82ae7d789ddff1cae1e999a923995715d5a31eb13eb2b9baa00572f56907a8",2388]
+["9acc22d4b0d5758bcf00da7dad9fbdec7286d13337d853377c78c0ca2118a1ce",853]
+["0cfa3b72e7cfa7e5025df96d5e482d266dec3abbc809e1adff9e7d520c85b4aa",665]
+["c909b4ea1296bf2678365eb769064c64804c292b30b9619abc6a83ffc1be7969",842]
+["7dc0d8323ab818a58fd53b0d41572ee6bb83c4a56e7bb4f53b897088987e30c9",2388]
+["b0953fb5de285cbd4acb8761da23f84d43fdbed4eeb3e2604e64d3c474efd9e6",853]
+["9a1c4dc1fe65030a3ea10f0eeb21430b88e607c326e860ae0f334d2c451135e3",665]
+["2ad7ac4dda48eb322b1ebb93fef9b72d1c713b98b584d22c80b5f12a663139fd",1257]
+["5ec96eda7b118c472adee38c1dbad932e738eac7c2671541a0bdaf8ceab344b2",2600]
+["ef2b1262adc85480a7b0ce2c996b83c0ff191215207e04e9a9097c512ad8f219",1283]
+["f71a7e1c4ffe3fc8baf7f91fc94e60462164ae5c07687b03c218d63e4f64b952",1650]
+["e1702c1be1a3ab433df952b4f0b068d8df029fbe91b345a0bb0fcfb7b852960e",1252]
+["e2588cac1b60bb21607d2f8f30b8348540fb15de6b97a5176066b7c77a16d674",2596]
+["3422ca35a76e630f73d498361946ccafc6ac14af8e3ee74bbd988009ca8776ab",1276]
+["63ebbf96257caef39131af4d18b91b252d0469b5d11abfacbc7255b1d6f949fb",1630]
+["29e00761068b84514f10bc97971d72a6d3c73dc451bf080371963a938da4929c",1252]
+["6279026f5f5796ed5912c260487ee9316c442d034af1624bd970bd53a4cc9782",2596]
+["7e661937ab1fec613708781de7c26fcd3a847dc562c477b69e66eaf36b504c57",1276]
+["42e29e6aa7e0ce625763316281db4370bb076dbf9be1368b5aacff9bddea93e8",1630]
+["f23e2c8e513b2f97deea57c2a1abab831f77e33d3d8bd389d264d809ad355bc8",845]
+["f3910ed1a97e5050a54104fdb301a5f858d479412a0fed0a2497ecf9760df9a3",2392]
+["4caf41e36992e0af9b8b9ca05d9ae616dcd86dc83cad0d283ec199eced2d98b2",858]
+["762ddb1b6ad9256660216588467647ae96859cbabe82a9fcc7b942ccf21d9d0a",673]
+["a373d849cd928d06ee8420a5fcf63ba0cfae44df1e1e8e2eb255e98b57c4e0db",1257]
+["4fa95e30e4b0dd2680baecba7d25af2f4d2251769241105600cde04d70067e5c",2600]
+["8641277d04ac5ce563fe7d46dd172aad6b4418e0345bcfef8553aa54ca990bc9",1283]
+["9311079adc7be85216d5952a5db9c3db8a433e660bf1ebe5a4dcd03cf61b671a",1650]
+["90a590ffbb77bc6e025ae84f96cc30d021922fafcb8b2fd392a65c761d095fdc",582]
+["235356c3a639b01fd3e138a7ce7657403d577996ef724b85e7b0c867b1931d48",2308]
+["aed1b39e9d8b45795510de7bfffe1fd473b1b4d3a4281d5cbce10d97be48b8fa",500]
+["efcac9591c39d4b4490accd6f5ec2fca7f1466de93314ea945c099f903d0d69c",398]
+["c3a87a3fa36b7416422e0adcc9667a294d6b021d2d2424bdd7de98c13e5ee571",634]
+["bd9dfdd7b07a1609f9b888b8b63ad8ed450d9fc300b64c54ab24fd33d9e91b64",2348]
+["8fd478b6f07e51ca86cda93e79df2401c1f03689c43c7401ddfa0ea9ef0aa80b",602]
+["105de3c95fb520c8e5ec1a4e60a19071161e33b1c73e1f30b3e208ed65f66885",560]
+["9491e72e8883b01bb87bbc5fef63d6bfc222b1a1371cf468de1b3e05dda602ab",652]
+["9010bd359330b910f9fee7faeb2b63605fa413f38bbefe9f789affe54c46f7c5",2372]
+["2078e57fe8e480aad22ba161562df06ab7af823920c3a00fc38d0f0d4845b0de",632]
+["b10109c0253521bb02467e57014bbe8e384d962d9ff64b65796f1c9d06456c0c",602]
+["fdb92551366d6e292964e788d8bc22229a9bde0d37322934ad4f7ff7d3eb4c7e",759]
+["5bae860faea61b695a3bbbc42e4eb5616806146a25e0da726c0db39d77c1d1c1",2420]
+["d2420124f9b347c2705f386ad37d1d1f5764d5ef5cb3a45afb83f541037acd05",753]
+["4ecf63f554f96bcc0241d9d55d35e53e020b8e596f7a8b408876a88b11cf46ad",684]
+["9c500db7b2a9202766f5c5635ce0cb73450ee381ebe24452cc68671d8eac6b54",741]
+["cb0df93fc173274728c2a15d8fce635516830f33a352ae7a76c54191e36a9c0e",2396]
+["cd5bd5c2b45877a781895d9bb897d27f44534a2e7d42b76ab93cea65abd926da",723]
+["cb163709ca9af159699adabb9729e123778b0010b24f648cd9d007cb92169326",642]
+["356c0e8350bf45e603db35d823e341e06db7a5b02cc8ec2ab5ef9e143ebad10c",570]
+["88a7bd71aabfc29113c6923b9e3df4132b6e5e2dbd11688df8bb4ee1af80d84e",2292]
+["5d6d900430d5a5d4bd46e8cb89a9c2426f9ee0efee4f1fae6b51cecc0137d5bd",480]
+["288ac1818722d2db667883719bff7b04d90c6cb09dd27202219d75ecfd664747",378]
+["220cd423bc6bbc0de2dc5558de1dbe18a25980ad159e7759c6a320acec93aaef",584]
+["70d9a968908da23922d8aebaf5f10b4794bd124c11f8987d56c315cb3961fd2f",2320]
+["d463f5ebe1ba638575cdfa3bf3749f5f5c45ab78c05f62d71df101b60b94e869",505]
+["73c2ea800ea05411f5be0e5f9f7163517df759f34767587d446e383d4ef006ca",391]
+["d8464e5220ba5ae6401d42fcaacb35c76629379c144126ab85aaeb5a60d78589",598]
+["aa00c29d79e493958111b44cd82434506a1bdbf2eff0e29bc4286d54c1db80b8",687]
+["9e4d148d4c7241294057f6a895d302374317743556daaf75e2f30cdba38509a4",602]
+["295121a5aa5acb522dee85cd668812a7f9c5e4bd78d8328bfbdecb9db96dbba4",709]
+["cdaaf2a335c9c7281b7fee67f4e19c2f200b8e6d0f6aaead0e749b66b7b2f465",577]
+["79cbbcc870995396325b8646212ffdbc0c9608224c2b2cf25f0594e7a62b2cde",646]
+["b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a",1464]
+["073e7e296a73053f7df3d748c729ec752641dfcfbfdd3b463e09580b9ec756b3",577]
+["2caee2113437b36b382eead66d696b102baf2164a35d8a78a040840498fa1103",646]
+["474f1663f448ce0812e38c045e3c3417dc1ed2490e945287e2e023b0bc37ee7f",613]
+["5d1d9b9bbbfe8b88de7312d920c3d2fddb5fe5e38c3f0506b89050e35e403a41",761]
+["ffec6e95059c8581d054df5feb2ebb0351bb744f4e751267cb064b5e98bab2a2",587]
+["e562c29c8d11e6f348c6b22c82a9edab569bd6dd80805c1861f26480f6fa8e8f",681]
+["fd0a59024284ef4b1add3c815fe34ec293c2e25ee1382fc7f017fefe1afda885",587]
+["17b73da018ba1b71c15e9bb4b0b3010c2b7bf17cdab9bac376c1bb6520f21841",681]
+["f99cec9a3edd1cbdedf904e9079b82446a9ae1c2f6152371eea4e33da89e1118",611]
+["18317315de34d7bf761beb5f7d103848441840ea8fbcc3ada569ce45c12a39a8",752]
+["1cb3e70113bafe0e506bbe624466814fba50e814ab8d99bb1c63f1b56c6d7192",640]
+["08e47e87f7e5eb3724d59c68437048a1580b0b5fbbcb2b4b4fcc9671016fd59f",2392]
+["3422da06c4cd4b91359af9177070fb6f4ad2f75b75481808ba64d66e1bf38e15",744]
+["65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486",663]
+["eb3c5c9cce1fd9f7fd176e1f45e5d2cd1330c1a9cba0e1ae60d23d826fc107b0",610]
+["d85ef4d6950dab24672b07c92d57aaea792299597a2f5c2e893ca31e1206db9c",2352]
+["95470f5d18532c827119cb844f4db5982b9d26850744521ade036bfdee7edd9f",694]
+["1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e",592]
+["1d9af5187986257762ca9c7cdf8973397c8f41b48f5d6c5500f0bbac437486b7",597]
+["20e2d3296a95482ede5385da1ddcf8035f8359138bf278f214f9879fbe2d2a93",740]
+["8c42565bee2f7ba75b1ed1b35861d632ef99520a7e39fc2a013264ff7e64309d",648]
+["01c699b8b263b3ce7c4e229366d048557aaa17f7b41827f533ace4742b124949",797]
+["b4f4090976c09f29cc8ef86880b74f779c625ea5d2159ab318b255c127248e02",813]
+["8bd0172c74d9336c80ad238ba346c2f99ce030aac62b645d660bd8cfb494b260",2404]
+["aec482b6c9395e13148fe1a509c0482271c049d397853c810c1b4b9c023aba54",597]
+["7e0ec8c1682f5bca0e146cc056e7b4f1418e68b677c42a4866f74ffd2beecf99",740]
+["00c255b7b4a60b60d312703c12f36a200e95259c11ac70d0fedb89dc1769e730",648]
+["90a92b78fa03c7d08a9b999f5814fed35beca1cc4577b28be18a98276bc8631d",797]
+["0fb816802017378a8e1572d6f1cd9a61d7822808558dc5d2f7c911d1ff86d6a7",813]
+["23b5cc88e55d3c68edec196c007e6778f4b66e18339f64d29762b9e44566aaa0",2404]
+["d96d7ee91fc7a711330f2784687cc8fa6b4d94cc0afe98c33eaa26b744dfb4bd",607]
+["e18f8052d5a90ac60b0639d318067f5fde1ae07e10026a5554f96377e378f03b",775]
+["75bfb6940e08282f4949f892f7bfdfdf1eca40fae787866873b80e5b1bbb9b5a",663]
+["e35e423e75a30c2a16e133c3ffd20a5724cbe7059529d9dcc51b3f4abb47ad9f",827]
+["847dce1f7a2ef9990fd4700b1c42f0e9395bf00f3a3fae7f4b10e6c079c608ee",858]
+["5154f3854a4763dbd98f308f4e3b9c7347feb506d44bc88655730e86860eb9ff",2424]
+["216b4e8bfb1b27152a8529b613cf7904a17ba5cf47772b6238b5026a329a9ca5",607]
+["c66112e043c7b53a7bba178d1c6b5f12e13f4fbb3b1b5f371db5006b1e409493",775]
+["c3166c1528ad4ca584086b75c3c80ca4ca344a0bd6cb8644ac4e27990260ee91",663]
+["9fc24c681ac109ea6a15dd2851b702c0f47acb1916e306144f9a91f738a59220",827]
+["81a4dae07b242f6c9350eab468dd19c34adf10f6a2bdbecdae2fae1dc8c32563",858]
+["206139660595ffac9f48bc21719d27d76a5d8d4161f975e175d70cfe9362b31f",2424]
+["263753098612490310474bba0436da53e5d62d085d7aa0def3d7278f26e13838",625]
+["875367f44c1dae8931919093134ed2419f470ac0b3d65bede4ece075eddd72b5",2368]
+["2b0b2d24381bfd730d59d64b3a95be3595ec33655d424e844a3cd4f8ad2140ac",719]
+["e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68",628]
+["005faf2d97aa20ba6f5a5ba4031ee0450b03884d87b9bb0ffe844fb5e66bc162",595]
+["6adc4e3668978726e6498b956c29a4814783d116e7c476f0954163990bab994a",2332]
+["e9d09073143827cbeee526326da569e64eaf17a0629919bd41de2feb96bbd479",669]
+["73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66",559]
+["58cf3ae5f92af815b446b89797c7ecdf316e9ca7276e6a6c3dd650544efe18f1",628]
+["8cc8b53cf286c1eb10112671002772e6e637b4b4aca784498084ac0acf5c32b4",2376]
+["fce24fc6bb6e1d5f1b80ced1d9f03b00abb0c2569dfc959674bb25094d16546b",724]
+["be00dc0e407a9f71d14c5b73a5a4c53b06e4430ad20f888e5595499b25c348ad",634]
+["5b1f46c77a0cae8fd09caca9d71942b2b56cbffa8f13e8c7735009b35171f11f",598]
+["473b39f7010f6bcc5227a87c1933ec82d43c1099db5114ba32617dd8c88ad37d",2336]
+["a8a5f1151abddfd39739e755bf6fab561c7a186f003418f7ef4cd88af2c0f550",674]
+["19695568a4fde5c5847f25e5103f5f4aecf6b96dd9f408028fee66222fe8c82f",565]
+["8796544f8382f208779de4b6fe84249d3d4cb1243b4b99fbf07077abce60b2f7",603]
+["14e192082cba8669e5f2974bda123e28e510fa46391191f78575521cee54e1c8",716]
+["9c2f3d93585fe50330d4eaf10e86a808b401473f63e089ff3c1e06ebf4349578",601]
+["28e88f129e29cc050b7e3f7314fa92d9c963d9120263af75cba7d40d83b42ce9",708]
+["dd4984067cbc3bf3af681a1cebd060db832c78563fa6cbdb7334043ead94fa88",640]
+["38b4d071dd256dbdfeb2a8dcc4db6ba79ebefd986a1ceefdea43f528f2b04fcf",2392]
+["5567bf9fefab9b696a3a748cc99397a5596f1ea6bdc2926dd63a862f8ea765cb",744]
+["277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672",663]
+["351f2b25e4553b576c9d2aef6a0192e852cbdcb1efa52df64536db8a317ba6db",610]
+["31f2426b205d1a334de74f3e8bc84a975eed86149821b8898879423090f87b4c",2352]
+["e6ba55b1a8f35487ad80d02420a3bc5db351cb348cf5c231011f37cf63c3fd94",694]
+["556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431",592]
+["c7a95bf8d86ce6b3a02edecc421928a13a2ee19022a28ec70a17e85eef400909",625]
+["e18e722fd9fbe1add226e570385db73369f8eb41008aea2e0cc0d7f2f08fd0c0",2368]
+["ab90272452b4b6d6b9e476f9cfa8f0d50494dfc9aa79774e9570e99af2f11f6c",719]
+["39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd",628]
+["d367d6b91562e1013acbb7994ae1c863349642241451d62d6a5054c39ea1c592",595]
+["41386e336e7ef55a5d07ea95c7f7189ea35b7b1e8c6db80bd3aed744431b60aa",2332]
+["3be0fc30dfcf6de2aa276bb0ca74525e43baba35795c20a8e7f5658c461193a2",669]
+["aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e",559]
+["8522436ee6768182efd6efc9e664095018a9f8fe017928b913069d523b83410a",627]
+["162a8dfc5540cce288753de78b6b82534ef8ad09bf39fecf329a4347cd894461",699]
+["8bf8ae698842269b54a406c065c1e2c6d979db10f0b99c6b1062e19d2f49d311",621]
+["947121ee51cd8d19b7ec5a0327bb2be7bbd41b335c83946d0312d21073404929",2344]
+["d52ad71c0f0c7f7cffea9cecf3b794310723f5e05a345f116994676b208ceb29",637]
+["e2d3112f44b78e0e57538f4a03c0071c49040e4386729694043f856fb75cedf1",709]
+["0e5b89f697a018c14087f539578a4c86a8f74aaa73c8c00915e26ae0dc5433cf",628]
+["b0f51843d39408c5cb8be51d20e923d7ef84d4087dee04642653ea9540f4e741",2344]
+["93affec1af57746cf1cd45d072ea5aa6b326fd8dddb07e7b230871e098d706d9",639]
+["83816652e3e39d38b3b8209a4e587d11daf4ebf63be9657895201c503d31fa69",723]
+["031078e13023b532576686886ce44c44da667c490b1b65d351f479d09ba99b65",653]
+["f135e5a07daad40e005a538961326a19528a58719544d1247b8b3f9d3836955c",2360]
+["9dd54095cda0320a47b8048cf2fa6b75de0bedf6ebabc8f1ce3eb36510248dc3",627]
+["040c9c4cc958c52dbe3798098012429d60b46c56f43bd36860c4cda403cc3749",699]
+["7ee8fb3ad35811dbcd2b10f3b04594162d8512edba373d39dec0a59f43c192e1",621]
+["a636b5db9e0aa78daeec002d41eecc2a1c8d32acbe9e3a8e3a37c8cae977440a",2344]
+["59937933a49611493fd1e9db8ac6a499afb13c6e2b83ee167646fd29e6e16edb",637]
+["2296588127de38ca27aac3f8112c1a378ae61624b40eaea25e44dfb3e260a840",709]
+["238adb9fe59e8580cd0331c2abab1f4ffa6c0284e2e0fa5f08710112b4acee7f",628]
+["b503f9ca080235e559f0201355c89b6694c139e7716d1e24772a4eb81cd68c37",2344]
+["71be9d8dcdcd814e02acbd30562aa12622210f15fbce8cb75863e2c83d360776",639]
+["d0937861d730e35b5d798b8dc0dd932b49218fc5211bb9ffd4de0a988cb6c131",723]
+["b637c344bf059e45cb7e4cc68b252c53c958feaff7785a36ce366b158deb6b20",653]
+["f1c78ab92bf469a2efe90b3c973ca7d807f4b2d6b8b809586c2697343352bcec",2360]
+["67938e0ff7de29f24f79b28af32957718eee51f0ca0dc5f04667f082776764bc",627]
+["7e4c80b75f0581cbce5841512403e565c267779b0d198346eac949875c04194a",699]
+["cd617a57a3b067677642222f6e426aa18ba8792ec4096f1dae43955bfc32dc2f",621]
+["a135ccb47b62b5e308da8d95f9245dae562b97971f2029228985a18c4ce87206",2344]
+["d0d20310387516b41c7022006c4da9008a412c858b05981d68138b2fa8490fb7",637]
+["9e3876d6156c9e09a493a890fd70d280dfe108d0f0277c78e4879f3e797410ae",709]
+["63fae2ec18f25acd594f17a51ac56e40d5f604a796c67c44d5aa5cdb49989937",628]
+["342d2436869d8229ca2bb2cb257e91c7a58339573c9de2d73671f01f28165855",2344]
+["bbc8b1768707c34c87d311770e4b52d23c02cda98956664724403acc2f605d4c",639]
+["b7a43d31f8ff02a9edc0384d2f17cb414cd9483e039b5dd9cec0431463f27d04",723]
+["934c55bf3941c3633d446ed986a87be2a577280bcd3f802bab42a55ddb39c8d9",653]
+["127ffce2c4ab35e6de98259b5aa8015897ef53fdbaf8ff0ec225ec036df3aaa8",2360]
+["1e8de4ea8c34065e173f1403803073f09debf08845f16492e98c821e48f16d98",627]
+["95c09777fc647cf8e22dd63f1374138b6d5fd481171d6a72fd162c8c4ed97976",699]
+["fb5eeaa27c5c51e1aeb73f7c32996daf7373c260321a02e9fd21fc36e6c400bc",621]
+["29d9923af4ee20099a3f5c3a678304bc20ffd5b99c2c35dbe1cf3067b1616da2",2344]
+["ca847dc0e911d074b69d705d2e06113b0ee746058fb1291429dabdbeb55afcb2",637]
+["5d53ff77dc530d3a9b6d364e5218f5273d9fe4dce2305ae45c1cdfc6db2dde40",709]
+["3ab368d685c79d616f505f8b1faecab19e28224317d80fc9322ebbc117ef5a9b",628]
+["1c751e09dd0780560454447c7069fc503fd70f1664c5739c8968cee0b7c4ec68",2344]
+["bb7e3315504210513d1cfbf114d9bce4d9950aefd2546c3051680d783b82c817",639]
+["6874681339d34def326726a0e7b1af8ae22b05eba2c46a8ce25d8490039de867",723]
+["65802af2fed9cafe64decad1ce728b0aa3749877468b9c3e7cf1488696d996e1",653]
+["8df892cafad32b33ccbe8cc9ca471c42daf0f4767433d6bec90e878f7d5b4ab8",2360]
+["6649ef68375a3768fb26c8d46b866e419d1d6c743b12fbc0e73f0d87f3eab971",630]
+["cb68628eaa6bdcdeb7b016d1901fb34dc4074bea8cbb62d98bc89159d74bd75e",2344]
+["4fedefc920b87220f78cd4627328e224dbd15fd7a730a4cfde6b56775158eb7a",697]
+["152cc7eb7f197015c8f506e87cfe23fafbf14a55cc7408dd23d2e0b1a43857a6",700]
+["fa8dadf005c24b8e6b5e5fceaa362e3f4091d7eac828aa081cfbecefd4783196",630]
+["a1d4c5d1328f16930824c1469774f43ce2ee0e4988fa7407e171b19457f5c6bc",729]
+["4031ac096add8e83e4859e828a70cf4b37a14c072c51f3b2a86ba81ab13381ff",1464]
+["ed2492f82b22daa9ea09fd1433a4391f64525f9e220a3d4f8b92b5b66f44a1f6",2372]
+["08642aad6a74a2e230ad2e41873a20e04b80482b32657a5ec19a2039ac118761",630]
+["5572960477954255ae2c69e244e690603749f158eccca414fc602f42f6e5469d",2344]
+["dc8d023f327a914b6a7cba164dffe8ab9bf572cba6dab7ed50886440a3684f3b",697]
+["d83569396cd54c6c1eb01a6d5c3829774c6ccb955152ba9646a6c84c9782f2dd",700]
+["a1175144b67092e8be0c1e90902e36efa916380c6df47b6ca20fa0926980c871",630]
+["20bf65971916e685f65728327122641bfa64ecb78e4a588318e04ffd7e5f0cae",729]
+["9f16476917feca1a62cee849a96a82201d1a22f0372414adee6916cf987584a2",2372]
+["91d8793d01f17e46ea497bbe7c2fde9ca1f9ac37684555b90f6211937ab3a5dc",616]
+["d49fd88882da3f8f18445ef13b2ce01f29ea0cb1b67aa26f57acadd6d1df5fa1",2320]
+["53ba89b2f05ca2f36ce4f98fc6880d42712cef275a683cbbadfcd1dbe9a319fa",676]
+["288a92fa248f36fbca74892ceb8ee4adfb98dbe5dbdc57ac499138934b36f10a",697]
+["082536b40d26a2c6a26acf6f91203d3f37789d6e32d4dd5181916176edcb8a3b",616]
+["b9e4eb40c383f18677af4ea73b1ca7c3ae0b0044a41125a6feacb245c04dbaf6",708]
+["e8f7fc4c64149d863dcc1963aef71ae889f994867001d5fbc2e4bd8da6bf899a",2348]
+["05914e4ff1fd2c49268239ae96c0f98c6706f3b91fc2c7bb4483e33a3b55bf7d",616]
+["6316ab2244a97a5f759c0a0f378f6a6782698a6d9993da980c7cbb7785c48ee3",2320]
+["a4397cb269f6aac55e09f6ecce33f347cba30bb01ff836eb6897e4a9707ab823",676]
+["dad945d465e64d3fe2350c76b705446312bd0d35f977f89176f6abab3ce04f14",697]
+["4acbe65608435da836cd361b3f8003f4f1d9ed41b81306e9d0964b7baad1f7d0",616]
+["2f6c5d6faf0136e8f77966868a275d4c0ccc9277f47f40137bd6023f869e67c3",708]
+["db7b6d0405d8bc81d88fb4eb48bc9f3f19be36c491b781bfbd657f1b9148fd3d",2348]
+["673aecd941acd309d26456a3391acaee0652ea7e2f788d1915c9e26ea8eadd01",815]
+["6558f3b3278dca58d79da0789f214245d753c4dac5cf25d29ccfee6afee1bcb4",823]
+["be10a258283c3a6bdc7bea420946bc426692b31e2da147ca02c0d28e9068b121",566]
+["4c560ca7fed53bae2111ddbeaeb47c28556e0b7aadf0b77adb06ac018b086ff1",558]
+["50492a17b07a71306ffa3195006d19168105bffe27ee4149448098640b3db8a9",832]
+["a791289c90e626bf52d9d62e28eef81e9e4d26b024bd7aeefe2e73876e65f20e",981]
+["a4bf03070850eebf9dd92d2b3d225b883033b94aec29b55bbf57561b4eabb774",1276]
+["9a45282486132d6aaf9f16197b0b214345786b1f6c35a070c9f57e2673a55fe2",2508]
+["c075d4d7cf9a01f85bfaff19c1ddba5a5effa900a1f928e6f66bba09a2e2705b",593]
+["36b0f5fefa2bb22b7ee713e976ab55ab18d1daa3c3f8982acfa375e0d5f5d378",646]
+["f8bc7e9d1865986dd97cc207a048972a8cf537fa680ade1f22df4e39b4270910",469]
+["774e4105e108aee30ecacd8ee0ba17042a28ad8522860ccd4f12f3bdc2ffcceb",2352]
+["844b1dd1fdc52e6c8cce26495b49a8d7b0ef3d72fe016df65bf10326036b706f",593]
+["38a1678676fb0ebd8ab12e835861b09418258fb7c994d44605435e429f8ccc95",646]
+["2e3fc43077b9f334300d3f2d7d6e6e99d74c9d4509e066abd9abcd677e11f418",469]
+["6265f16056ce2e3d571eff1ebda796c31797e4319cda3b108b5d0c581c811218",2352]
+["cd014196c7a090afe7672c07025ef40c5f9ee5f830916ee1a0bba2b460e4da19",563]
+["21a1d2fe684434a6c71b2454ff0a26463002def6ab7c09e00252a70d9770b7df",640]
+["751234f422c60e4655455d4e48f91f1f44c6e7949e9a2c981326e35ab4904b72",573]
+["e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2",592]
+["7f45f94adb49931b280a6bc529f3f2205e486d66f6864d74b4cb93bceb11f499",581]
+["8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0",617]
+["b040fc36d0dd88bd0fc7e9f95526e9ad266f9ca87f0ccdbece0f441fd9ec676f",619]
+["b922768f98b5a1247e512037662ddbb34671d77a2572f75fd95a6606e42cc7b3",508]
+["593c47a22026bb69ca054ca7925ccce9993f6309f97c8953b9e9186b922f05e6",678]
+["03e867af2f62c91978435160448d2acbf88a2555a2281d35da2abcb0f781f081",630]
+["0184d5fd01b810f78fb9a962697abac4f80d183832d032f4b86738824f404881",670]
+["16f4a64a101980552ad771ca7f52792285d09d09c30b818e4e94f1b738366c97",610]
+["3bac766ddc3ce7d6c72f1d58fc08d004e23f5470b3248752050eded880158a00",606]
+["5c80067aa67c4272f34796bca90e02033a6adfc787b9e3d028cd264923b01d25",474]
+["6352e976695a7c944eda2983ddfed5f4ccf3e1e48b69d11f042f2268bbb7fcd3",701]
+["3b06d96d10079e191533c19db0c65e938f8ec65e5c3dfc26f6a5fed9d77a9e49",2472]
+["0d5cb542d1158eaba166d1bbceba2dbc16287a8dcbbb461bb36e7d6594254134",768]
+["c51c79f01c3dae0fcac4b6756077188d947db60aa19374a3627debabb137021e",714]
+["9da8433271f8b366555eef3448a04abc4c32797900a47adcd336d29c6457374a",689]
+["27eae0d69e82a597195f6bb7d03917a9efc7e02c968f5d10c3024cb00d593df3",2456]
+["73ebd610ed21c9de57892e75b5812095c572b3b461e980803f7c4bb368b1bcee",748]
+["60d0d055f43010f602b6857dc92b4396a9b6c6bc43517baecde3ebc8bc445121",686]
+["b177183d0f5ae668e91948ccc5aad8a9e2aa106b21920d70f321ea7018338204",689]
+["71c526f7b891c83678a69df0184ce432c709d59c1c48fa661a0e6eb3a0d7fa6c",2456]
+["404f9763b66d6b02b3845e74f4445ff7294245221b738e2b3e3c3b0cf6e9af33",748]
+["e6230d39cb4d5879752791ad6f16df6c43465f98f7a5d904e949eb1468a9fff8",686]
+["2616ea656274f270cb6b8027df629aff20327352e473f8ea18a27dd4614d2fc2",692]
+["e804edc9e5d9748ef44a0b5d76ce078cd2d4daac3d3df67e1b8c936861be01d5",2464]
+["15e3e023329cbc869794c571ee16fc0f3f58a1735d410667dde3074055004882",757]
+["6212ed77c02ad417fe4b3fd49c1724b14e162c70064acf4b82993f28410a02b1",704]
+["7d3294cd2ba510fd3dd0b6508c574ae99449ae38497ae9d28be555a40b4e9d8f",654]
+["615a3cb194c76980f9fa91c738b2af3270bec98fc9c8d4820c762771f0e976a9",2404]
+["560de32ab6e1b0990ec92de8e90ec72b1acfa105584102297936a94c99c78ecd",642]
+["f914303c94fda02ad91a4bb9e13cfdb3966d29d9be21985fa2f35e1b2ea752c0",519]
+["d837e56d583371edb3a52c19a83fb970459b8ee682582554b11b42195b145143",642]
+["8afa5d76b2c1dc92a72f6582b877e4068875acfe30b526c338531e3ee4ed34c8",2388]
+["281cdd6e4ad8587d7ca1d0d43a4f3bb4fffe441c57ae99dce7471afac4d133a1",622]
+["1a482f767e8a8f6c02bd5b04418ee374e605885d12598bf13183f56c729b5932",499]
+["e0b1a896fac2e26d44db30bcc42d77c324385cbf879193fcde2b7237d7973a0f",642]
+["ffff500303eba3ee37c6fbf04433603346ea5cbe8561b029367056f41c2fbf3d",2388]
+["1027cf2afe608786567431cfa85491de9c5d0f6b64eff974670a5f708a672509",622]
+["dcea04f60c6e0894ff437978e8bc5a1541d22c0a376a660bbfcce687fa4efa70",499]
+["bf49b1b668e7adbe403dcd1c161149cf86ec1a602676d8b29844fd3b9f79639b",645]
+["54c13ab9b7315493eba690d49bee6ab1b341038832627c6de8ec2030d2200dce",2396]
+["b35a8e565086de6cdcf987399b9949df62afaac763dea773f5d34399afdc63ef",631]
+["29ca629ada6d2d61ab1b9e5dcf19c34d802fa2b75ce51f987d4255c325cf8d9b",511]
+["0d8237177f69e379781a86e4b14a7bdfa89317ac147c96b30591a3c74a4e7511",614]
+["55c43ff116522660c3299c242db96c9f94457cdb4e5fa9d6faa9ab34a79fa940",755]
+["e5cb40ca0492cf5a54281e196295221bd9d6b220d6396eaf18aa039309724a3c",612]
+["597ae289731d7ddf1b997a1d0a0d9bc9ecf1dcfd325e6d7d3ccad5386ba71d4c",745]
+["0ec8ba6786e5c56e69919717e6701a452444be03602ed6b642f9ebe558e61223",612]
+["048fed4feae5c84a5d1998351b884cb19221d46dc44dbdec2365aa7f1f161398",612]
+["b5a61833128fe612fa44231561c3abcf4dd6cefcf28ecfb050997f64515a8dcb",755]
+["86051e7b94a658d9b74ae2d9e41744ed829c79d83bf17d87590129d0b5576204",620]
+["8bb2b115bb790c531382367dfd3a34b62fd6c621c845914ce0626bf96047a30d",785]
+["b5cf90975d2121f35cea3bcf489458d40e4cb25011fd58afeb7bd791aab75108",620]
+["f5a74a755c546dc381c8e34faa9ccb1f2037e54c621625ea8a78517d7af049b2",620]
+["ca98c781356b9fead911df5132c8aca93ae52dffd3ab36f7a25509ee0278c621",795]
+["8a8e603a300127bf1457b6cb19e3030ff9b06cbbe7d78414c4bc3efd28cbeaac",630]
+["163b27805dfdf9ac4e08a70d876363df62c83542b8849d83960dd06c89f1d202",836]
+["f1401263d3f7f3ed68ff34425b4bfe8dfeb90e3c50215403b80502d2adfcf164",628]
+["c792bb620eb963b3868726979893ee4cd9bbdf9419f99541a8e9ad2e73819296",825]
+["b30fceeb50db370f9904b41f4ae04c27e0c9f805cff2e484a137db41944f66af",628]
+["4999b0e5b9d37b1a030b56e1a9cac4021fed6b9953284272cd580030c7ce9292",624]
+["5f764120fc637a78f11b9468ceb47fb7f2cebba5435269e0dd12091843128742",805]
+["616607ec574d9d178ac73f55f4f278bc43b5649aba86360cdf4ec02efc434da0",622]
+["f8416d4267c73b96623bc6924b604970da4371d53144f5d6569bd33c1883ec9f",795]
+["254cee509b59d8b2237a095378b3d3f2eedf0d65bfd19f7b684572b409919d82",660]
+["06ced826c93c895e7c3668027e79845b8c63973d26cac1c438e0a90ce2231e4c",758]
+["71d91c4eca79e1a3a25a9e7637650fc0ef19803c1289d6552a93af8896697a75",713]
+["71de23ae8375fc0e5d9925e6cee0acc16201132a7e2edc268263b275e845f6a5",2352]
+["f72f27aeb2edf51eca39bd8cded3d57483778c6b0f0fa98d9d40cf4ddae84095",673]
+["57fe66809e74c65a88f6ba98979b61476c248cd9cb41d4f52fc73b45658374ac",771]
+["ecdea9a5ae66e23491eb9b65df93c2b7921e65327b94efab2cc2ca3bc1051f37",722]
+["5ee75ecedb5f2628cf7b323f4dc72ff478c60860f28869d90779444691c1b413",2352]
+["f408b6b0f4a766e1e90d4f6e9efe62e02bd12914ed5092d8b9e0063a3ba19831",672]
+["72e39917ce348cc80766cb6383e76b6a4e0f61943eff8c3f472ed6db76ce7dd1",782]
+["f3066e4b0b4e0fe4966fc212fe688d4406aa25111f409cf1dd5dc74d5a6539cf",753]
+["aa5250fe7feadfe4fc65d1ebd8c82c50d3f0ee10b21acba2c82c58cde35be8e5",2368]
+["bf86c1e4a177a6e171cfb63bf0643b0f1de847121d84c90872f0fb3fdcf15700",645]
+["7c42c71362eb9f60777be3424bb4d76625adccd8f2e68faab6f1f535d0ccbffe",833]
+["3fa68de482228fa742ffb503168358678b5819e68d540066fb5ac683530b2c6c",637]
+["111c17ce3683d228a1396c7d573e537cb7b5f97bd8529b52967ebf7395921304",792]
+["c459d842c4251c3fc55501686a06b5a23671cf98e550d490e33ce92baf4aced9",660]
+["693c86b919af07ba3bfd039c53ccd268cef5e2b1a9931c43dd5cf576a1536ecd",758]
+["4f489b8b01d411925ffa814d55b822a5e1c6852c29e556a97c38b17661a3301b",713]
+["b39161bc7b4609ab001dc66006ae600d0a2cf45c3935f6cf149be88e36c4f22b",2352]
+["b838a9348b162952974d339e3165e894c3c3336649a5d12a9cf625aedacffefc",673]
+["be36b2139aa1e142774987f4516de77486385ff9afc7d17ccef0b26fb669ea0d",771]
+["78945e7bb5865e17fbcc96d469e22913341ff448696b1f7881502b77c95612a7",722]
+["618afe50a7b77a320a924febd3bd6764c0d7b0643185edba25f9fa9a7919c9ab",2352]
+["5f7262f0fe02204aba327cb2e5674453ce93c43787ac0798f18dee0860cec3a1",672]
+["43bfa5bb4d3c7eaaedb2844d598709faf144ac985bf4deb443672db9a949aaa5",782]
+["4c0cfb3c14794e97d42b9d762572079bcc4abbe03a3c22521a76e70d5525c09e",753]
+["45233354193cb4989322d69457dc7a816039acd71695763339be5b2e98912caf",2368]
+["16621bc22c42b9565b9e32098feac991c1a9d9fcce237b04d5ec13b5c67ca72d",645]
+["7d7b550740fe20e5755b0d3d07d683130adfddda51ecea04c47aabef5092a4ef",833]
+["445b63d58486f5c0bfac692d681b176c43c7316caeb6216c03258afc76cbbf65",637]
+["9f1b394b6fb761d83efbf4055f8a12a83a07e2eb97ce5c11354abdfe90cfc83e",792]
+["8ce779fb012bd9e8b606b0553259d6d6aec5783f50ee1a0596d3c3cdd59be167",660]
+["b7f707e7bbbcc2ec7948a42fbaaa6d560b5b0a1b0d7af30458f1f730f6baa295",758]
+["a0d214f5445f83caa8e7f0ebbf3064a16aad86343bf14bb6cd9690d8e146c839",713]
+["d74992a91dbd1a467245af7ef274d5e72810584454a521f65f35fe49e7c860b6",2352]
+["38f983eb809f87fec24c23e9e06aa202aaeacb24e6c926da0533fe5f6f6ca0bf",673]
+["0ed8d33d350ff97722c01defbbd087a5b70d482659b3a8195a73fc1b524d36f5",771]
+["addc51e966429beb242867def67d9a9ff7dd16c632ff425df146d2c72a95b253",722]
+["042a1e7cb5a8cd60817625da628855135a7059047981f3278413556566f3f193",2352]
+["4e33ac0d1e702ee9204e1eebde9b8a3f334983ec7c31ba58f47816a1d8188ae5",672]
+["cdb71bc0e1b861c2467a3d5dd9324ce81de22f2394a790387da89ed923775a0d",782]
+["28356f8d85dae217a879c201bf0c6144af812d102d2e0474acad6f01fcceb37e",753]
+["65daf722e6c163175d93e8064e9fba20598196e32fcbf0f546e5bd31b59ac0de",2368]
+["4e274e00da276c5c2bda9b532c7a0dc90aa8f9e3baac4b68cf89bc6ac55ea959",645]
+["cc1c1a66fd9a9e065b828e289270718dba6407521d57e3830053bea1ab84ceb3",833]
+["db5629ec6e1d3f0c934406103bc8f2d4f1aab5f86085a48c5d45c261b7386090",637]
+["a6951b03ff21f032bbbc16aad21f0e8b086816542bdb57ab46581c06ee65c3a7",792]
+["6259cc0de3184a6f6721756be286ed97bb9e47eb76b543834295c08353bb2209",660]
+["6e4cf5e03b17f0c736662ee0851568de6797e063db5b534254e9ada6917fe70a",758]
+["a6e8b5ec4fa38713f378414770e6ab342ab4d0a19578fe3bd165ed42bccd7e13",713]
+["cc50b23ad2760dda43a5929502bfb1d49b174253b6746b9506cf42914d4f3442",2352]
+["78eca954604533d70893cdf6f58800bd6b0815e877bdc1ee586858713f374189",673]
+["b5e13098238a128ba4108e4a6b07f831d42e0a08c66ee77fcd73271cf5a7c139",771]
+["2bbd25fe26219af71fae7a4f1a04969acb2045aa8cff9fe18baf392f0f9833a6",722]
+["bb70297c94e7249a40efc84c0228cbe927b81ad33f86fa612bc27aad79cf4c89",2352]
+["c81241afef0cc347e9eb9c5af353e46c2b00a9b0a4af573dada0f7b310af2896",672]
+["f344e16d6689a33a6b3568cc7d8d35019d9a665d08c853059a1f0585efada6f6",782]
+["f42d1d65848ab217df043b829da3a9c46519c5268305d670cf07d38b7bb475b7",753]
+["18e3d316fe671f5e85ae9b04f8fcdbd4dc58bf21de9097a146916fac5d98e912",2368]
+["1964d4b8ad55b0bdbf53ad441ddf097fc06e0a129d7f45cc01221885018657a8",645]
+["95fd5f04fdd6d06f672efdbf0de55094a48d8f6aff4d4600cb6a59fab4a04fd9",833]
+["970ed33ade14ab659f7053b86bc22c5e413eb0b79c106fba2c9e14a871f69666",637]
+["030364a2ef2b1f0f80888f41e6b1a698c15857cbb20fd3d40eb75b5f9afb0dc1",792]
+["8afd7afe2078062ebaae56f63c33aa7b4c6f2e8dc18d9885a23c470dd659670d",620]
+["b3abbaf51c02f3f0e8e00d62b84b032ac32855a0ee10323efd0939452b1eb325",785]
+["eac550fee764aeee730627005f8b69d7ffc03486bb352af567aa5ebcb4dec3d5",620]
+["b2e583109f091afe5781040801480f19b98a6f196aa0f5b64c81cc6887150c0c",605]
+["8670273ca215fc86b9ca6cbb7107807e704a5b9b02ed3587deeb671f650dd517",674]
+["25bc5a6521e04a5b649cb7379e1f9e78d71f28d4379dab75c627c1fc58f7f8bc",492]
+["58adc15161a3ccbaae242c8a7b5be1e19c13e1f0279654875a4c4ae10e7ce77f",2364]
+["6ddd4d06f46d1995f487292fb27830f5684792c096f79b1e2caba99086dfc43e",616]
+["1209213c2dd78fd7c3057291e87074ae76b46ab4f9553ca3b38dad233a66347d",2356]
+["999b29a11ef661106dc86e44af7c232efa7bba7ca6df97d52a8cb7bd1c12d4ee",558]
+["97504b6bc748ddc18d1e82bded39ecfefb288bf088e5afeca6c519f3a4f900cb",457]
+["f5a8569166d90417449feb1cac3254d188628ac5400449d54bbeb6e9afd112fb",577]
+["c99c2f29870eb6c1e507f161ce58034db4c794d757e5c2d7cac57378dc51f862",2300]
+["dd1b6b51138758ddbdf5e6c1da306ce2d481078dbe1941ee76742350737bf7b8",493]
+["9b54ea0566c66befa6ff9e703c91eba251fa5c0fb9eb477a07c391fd9556045c",392]
+["87c5faf1d1d7e38475f7099f959574c75a5344560a25da1097c93d980c8b0951",588]
+["a8c057aaf271c5beae19c2ba3eedb2ba96c42668c304da0e15dfdc1e0c615708",2336]
+["f8e8dfb8664dd6d2d3623c5e4301582e10ac5b46673a0c25d775e68383303e41",641]
+["2789580e18b37e2960cc435b66338812d7ea8e3887e40fdabf5903e8fba1d44b",445]
+["b56430303bc1edfe02bbdf2019688dac12d84e67b38745bb1c75d72305f20f12",588]
+["1f4018c64e3a1e93608c5304320c85f266719bd44c6c88a9f52b108e4fda18e3",2336]
+["ae0bcf11fe93e59582f39a8791336016e0b10acc130d99770ad0376e15dd7e6f",635]
+["959ca735e7c1f17e8ef508d72fd87aad8888ef65afdf7529000374d4ef63d39d",445]
+["dff4fd0decaccc4d27f4002f8a9a08fab9efc22531f753b125e7d54ba7a2cbce",588]
+["fd8f78f2f19d83797ec5949338fff65c7f773f51e88472a5475a125a909b867e",2336]
+["bc4605e9f8b4eda55436ed53d9746df168a3677cdc9d7eca433b52696de2ea87",635]
+["fcfdfc8f382120260e559a6fdbfa9cbb4b39b4d8cdb337b4739491b2a0f98622",445]
+["6458e14511936ba78a2a608cb4fc3df9e993754e4acf59f3abf8081cd7e08df0",544]
+["db526bddd92f35cf88a6651ad40bc526116438eb547b9e93e6b5b83aac385948",506]
+["114276b7560b36fc777b914f012193771420998e2df6eee9a5e332edeb27d80c",581]
+["4dae2fa1239806154c0d636219b9d5e85981a74f58e3d57770bac276169ba6f0",2316]
+["8d3ce0f37e2869cf95bc9e4062d8829a0c3c0833e1c15c3adfb2692f78450c13",619]
+["3e3dcaf9017a6f4cf71ca0902d815282ccfbcf15aff4dad8c9a612e41195d028",434]
+["fa23ce2d114f09e2a786c4b1f80e2c57ca64c74f099f0aee73dc7bc17f57300b",595]
+["5d31285942f2b6911c4aa289096775babcce9377276287e288e2f4a52ad2d844",2332]
+["1dcde8d3871fa95b495a5b0f014becfd052378b4f105223d574076a40bab0733",624]
+["7b00bbe7387e067f30131c854867f994d1b99265c8d56a3bcfbd7fad0dc2d8ab",451]
+["60f998c876d1976e58260bad2cc2e0ce8b0a116cc634f8b3321c07094f4ff385",595]
+["c61843cb5299dfb2d2d932cc818bdf7b99e4b6b04780e1e9f60d2718ed87130d",2332]
+["1871a11ca3f18cd777cdee15bc75b435da8b290d7f8cde92299b8f404e0f2ed2",624]
+["4aa4f7d294e01dfef9a8a779a2344f5bf9f144888474723e2735b6997e00e797",451]
+["e7171ef370886c40389baa3edd2a2345d262387f16544ccd27d46a71a4263b13",595]
+["afec0ea3803df1295d6f1820f5ef1b04a58a36f53ddc55076aa78890dfab35d4",2332]
+["af1885ec6730154a18f56eb4e3bf78f8ab5e575e72c393be71406551ea5965d6",624]
+["f56414a947bcfb9973d1eb5b8301c0c5b2bd301a458df5115d8db89e90076401",451]
+["9478ac8863f1ada98c67af5e62f62e63e08a665c16a5329e42d31e735d60f256",595]
+["2b77a65b3744874fd5690e46bf6d320fa261744e4c480b5f53cff6e93deb2409",2332]
+["32e529becb7d3deafa64a975465477617e89e5257f4d824b2c65f2f2ff2ecb3a",624]
+["c096e3837ac05b2575b9478a3d174016c28e5cbd4e4c84b1dca0fbe936ab71bf",451]
+["2c17919351d86176256e9ca2412eba264d156e266348db07a90ab6290d480878",595]
+["75ae36daadf051ab40cef6c448ff1754b87e0e511e0963414fbf9fbe0013b590",2332]
+["9979927eb3211d98435a02bc45d82880a4aa555744e59984af4547e32041e759",624]
+["72bee2ea12d8bd26c09a5ff84eefce66cee7e4055c8b41dd29c7537b5169d74d",451]
+["a0aef3a4733868377f7851dc47d6e8567ce57b1c2e5de6b05048156a968bb69f",794]
+["5167f9b5a2b120a330b15282247fcc7d11453867057bad7aaea86d8e95bfcc94",2420]
+["f2f0a7b92fa03903fbab495309dacaceb4d3152154efae751c291aa6c3c947c7",833]
+["236e9bd4dc340319b24a2adebf8b7f83f23046e44ce6ad701045a104fe90cfec",775]
+["c2f5b17c9c4ea66d8cbcef5fc63803c98613469bf1aa6d477271417878630379",794]
+["e6d2e421c5e40be07aaa0e5d63682ed19ea62085d288a9ac12f9418df8494ca5",775]
+["5db06b911c8d44240470b18d351cb863cb7582b3314ec3ff28717aec38c32fdd",2240]
+["15ba0d82c34fe491d798bad56bbdb84a69fbb87d5e7e8fafadaa6dee85ea8537",794]
+["d0c334d0c22d1837bb3a6dad381f27088b88b66a511378aa96b55774c0af91e9",775]
+["a8017dc9002cbb465c182a8d7c5ca85e1faad65618a24efc453d6b60495b434b",2240]
+["e21d50217cb3db21ee01e3bb28476c8ac5bb12a8af3a9ca0eb3062efa68134d1",794]
+["c33bc20744e6e6d120611d4db30ab1d643ab8619ab5d9d67004527ef199ca2bb",775]
+["ac8545eaa1b13f66bff879668cd7bf84ebc48c8be2602cce6fcade9303c300c5",2240]
+["fb44a60385c83846193d2954a9bf98df449d18c4274ad9a66d839649b7804029",794]
+["550ca831669a05424cebcdc37a58864e59ecce426e1a4247732393e1b2d497f2",775]
+["bdfeb2a304e9bf915bee1a460397c878078435a02745d0b194e959408554097c",2504]
+["764338c5c541c822995d62c879fd857ef9a00935028f71cd66eed721d2462bab",794]
+["59eeb96730d52914daf2758cdba5d9a626be88918727a18deef86146f0b9b65f",775]
+["1fc7f6d3632ddac58f72ef9c01fe8750c9e55a92fab1bc034bbacec0918be5da",2504]
+["86efdef2488cabd54e0e8b168d527e4c307ebdfc9aca8cd003ee128475b62a56",794]
+["0fd0d36bdf8ad64475d4d99367438e88f67a83bd290c98fe7f8af8f86576db2a",775]
+["dc2453a1298c2e8c20a23f0835fb64fb00ff66ea75fd468f5d0973e763f08741",2240]
+["90760a417bfa8d9fabc0907be94ef4d1fdb72ce48ed6caca496cf141648ff0a8",794]
+["d03c19f772475a47190da1a857f286d5ffa08d7f8090dbe7ce4d35dd25c84ec1",775]
+["ca980ee914906d30713cf21fb303b36ab9266b98c76fc0637ac71dee040e294d",2240]
+["327ca2f14a6bf71dca67d4b9e97eb4dc75125ad431f2367b9fb9ced000d8a088",794]
+["7dc8a9eb460aa84fcf987202f49a1d2f82a6d0f6b66188682df028a3580503fa",775]
+["f5606e8f4e57014d796f3abb8ead83732155ab7ca407a76ace9175a697491f0b",2240]
+["2910fac62e65dc721c5daba263ec1da212486ad8558280275f6724c6f354b983",794]
+["f4c1c54d490e3d5ee9277b988a55a5dee3ad4ad4ab19406ebde3748dd97297ee",775]
+["0fd90a71f214fb9fbaa285c37f1bb563627e58cbe0a95a4aef5b2f7224de989a",2240]
+["0aeb777c6f45f5f6e3a4b88e33b5251a4213fdb9e9b024830aae7183e51d8489",833]
+["0a7953b484f97df16412ff81bc63eb3baca00a5b681c333fad812268b3bcb851",2472]
+["08d08daaf08c78fdf44988ea390ab2f739c213cc0df5f87617e31975a61bafb1",898]
+["d792857ec5dc67822736c63284f2e3c4c63d2844ffadf8fcee3b2954f796be6b",866]
+["2e0a514f026f08b080d26dae78c08da03df1c9ca97dc5c4848d5b155af769c8e",833]
+["97d5aa12284f82b0a05d8660ab60867c003108375e7612ec4945d62e45aee198",866]
+["bac52bc0fd3d742c1ba057701e3a24c3d3aad9aadefb1931ec24a870937f690c",833]
+["33788c0a200f4afc3d50b771d15e4bc645c0e679c62e6f3e07fe433f22a59f6a",866]
+["cad6bad1d66f59387d78ab811e6db81acba13607f10089fa4e2ab45d0fd4e762",833]
+["a73922a3de289cb07b0c752d5de3fd5caf9445ca0c5252459a3e5f50f037204f",866]
+["b35666be64f1d99b6029743e89cc8df86fb1056d75feffc9f365bece047f72c9",833]
+["1188729056b2eb45dce69f027cf81303f3b4326b0373f5dd6157150bc18a9780",866]
+["79f4996822af991aa41bfe049fb04ab04906bad48519b4ab934f5339860c020f",833]
+["fe2b4247f9a2d3e73849bf40e66605dd03a3d0f94c2afc003d979e858c89dd3f",866]
+["5dbcba012894653a180efa2277f4cf37a0127e34c2a00f7a28e749e76647ebad",833]
+["5e466b05ec12eebfd9ce43cbb91a2c043e275cef52c7bedded5bca31ae9cd968",866]
+["39398803d584ef0a60a24766a0fdad587bc4e576c4c9275f223c00afb8c2061c",833]
+["0e5cd2b4c3217aa8b022525e75f6a43ea9b9e36c200629fcd80328e5a1b5e1cf",866]
+["537b082fee0c59a833eb2b7046c1024ac9906e947b735aa9e92249a538690003",833]
+["12fb837c27b11b46be2a2237058fc4ce57e8bb12a4bb53be272355e4bbe77b6d",866]
+["9952af2741dd72722be5c03380c694980d6de9080339da9597632e5a3057c40f",833]
+["94c188d61eb655845a646df342b041e7b9a551557f46e21f558f2e63e8c06227",866]
+["a13fa339a1622df79af4cb7b0cd3c77797910cd8315b476a1a24f4442de166c8",833]
+["d2d803f4a522185cd452257873abf59502e865f470d3a6901ff27abbec779340",2472]
+["564fd625a02ac360a0564b3f0dcb041b8dd70315428e4f925c6514d449c6306e",898]
+["52ba2b98b8189c8bada18ad1feceb0fa6b2f4aa82f54ff94315797476e9b8879",866]
+["582e6e0ec1001f177539cb560f41aebc8b6fa236937f7277700efdb0c86ecafb",833]
+["6c812357f533c634bfebdb263841facb0c8c84e050a0790e33f7fa53bef810a8",866]
+["c531b9be0e245caedc07fb415df69009a88f9d65ef97b20fb0edbbeceb22e6b2",833]
+["f264f4242b2ebce62510611a4c6062f380fd821f484d62d9f1edce47809ffc25",866]
+["3b9821ef51f2003577f4740e9199550bff9fb89ec44fec72dfee85b88cbaad9f",833]
+["50c46290539437b18dcb7f45216ea7f0373c961a9a972a140b46cafcbd3f8af6",866]
+["f692d82f7c874e1cb7f0f74d8b1467f337759df3f3100475f5d0361782630273",833]
+["0c00d21a0095105b6e18f4439cdf7ebd9095952cea1132e0f2a96e040cd54ae9",866]
+["26e90d9c836858c6abafbc739981556ac4e0864615708a330f5c0dc8123a7165",833]
+["4b79c09596e2d2b4766b34693d22d8ecc8e6ff9506eb465c95be0bd6495b588e",866]
+["6360512897ab83fbb464265e6bd77472a0d2c24353817590e7fee8d6c2b222cc",833]
+["7a4b4bd4834cc892bbe452b0a89b81f676bf345fb38b4a28c8ee970ca367aab7",866]
+["fdf743330f5a8b49b1703f380cac59b69a024b2da3305098db710dd55ef7a96a",833]
+["7de0d78bd58cbaf9d6ab1be6d6ddd441bf1f1f58c79ff40adb937853ad361ea6",866]
+["032536cb86dd61d9da9116176fb7e6d746bcac37fc18420a390bcb35c9d2af91",833]
+["6c2b5f5f86b67ef2178292c05e0c6346f93c37c11541dcb5802b3f67e24d4bc2",866]
+["e2f49c041fa8253f6ada57a8f1771a2d5219582f320935dc48f1e4814fcc70eb",833]
+["356713dae5530da86f6620d2dca0f27a512afedae6561e15be89abe2784d69a5",866]
+["9da4195c7ad3dc8dd9c7098ef6fc70b0018f12952eb406a0f82206b8d4ffa71b",834]
+["d7455f9f067d74ef27d5f47753f58e61e87cd4bffff4e22104962d58118f6898",2424]
+["9daf17abb0b84889323c34096f9d66ea5e7c0eef8c63a63fae97ac8072d98878",873]
+["a341849b22e87e11f6e55c79f769ff347ec9fb8627b15a060b3a51429b93f3fc",863]
+["7b8bc71d713a3e7300247f8b54bb04ab11c1b1cb783661353803b0b68adfec2d",834]
+["a79b94e3815956e1ac24009d8b5a3f76f7a91c49636b252f59fa3ef48cdd89b3",863]
+["e734f5cd92789510a8a3f1bd81d8e4ca6ed09f65b38161a3d92a66d40d09b4ec",834]
+["31ec3b13640555920d8826159017e656caa49943c61ce556d589914859e36493",863]
+["9c00d069af1f3bb80f9de681e6b3396d1d3eb5fb6946e49ac63b78b79f3dba2e",834]
+["6418aa5f5c56d226bfcc143c300a9fbe64613038feae22f9068f413011e2f660",863]
+["9ae1229a74a179abfc8252808f08db065b34882aa7186ec1a93e5986ceb736b9",834]
+["8029f86602c2c395180620089ce455cf3d3254c3617d31311cd3658f7bf9ea21",863]
+["79d3451a4c625f68bca033f9ee83d2308bdc362b2b5370978d8927dd65c77712",834]
+["1b0434c3adb0f109c51f00070aca27969eecefa4fc8707406a4064613a8e5ad3",863]
+["8eed6ff6e2765800cc56421b5f996906c8533ad866303d660617e27d86b2bfa2",834]
+["15e22d40d6627692d3480d1617fb8ef2dc9681c3a8230d86de9d65360e351143",863]
+["cc606db72055a69e770c67acc20bdc5ea5e4309ac57506ee8cd6827716b0898e",834]
+["65e2aad3a1017168bd0ba274e4c4a981777c2da59a8c6ed2c5a4586ab1c9ca2b",863]
+["f12cddb7ce7166ff6a34de0ad2b0f1a35025550b9ac2fc845d06695573f896dd",834]
+["7a330bac60036ec1d82843a22e781d5aca9f45da377eb77e560381d91747d9ad",863]
+["9471a1597e6506ccab976ce06850f6360bfaa48710462ffdaa31f3afc0654e7c",834]
+["64a15b2d833c815366287e81bde27528a7ed92a806ac126b6f22858a43681156",863]
+["6ef7f365c20cde88cdc1c81828fcca646dae0772c1fc9e98a0c1b70372e1519e",873]
+["d288eda9c86148e5aa763c9f948a14e19d0a1b68bd678bbf22f619998fdb4d15",2476]
+["fd87bedf1a1d486a7adb9c173a36accffd035c5eac6d6494da910f96477b0289",938]
+["25dce38f2b571af1dd09098849f62e96ef02c79dcf100e1bd01ff6195e8fa46d",954]
+["2a52f1977217e5c232c8cb9862c847d97097cb050706b88e9a4a956401e97d9c",873]
+["82a7159a958483aa2773ed91b81b42dc4c6d214a68958848998a426627ff5162",954]
+["a41a341ae18210afaad5ce87443872a907e3c73fd283a6e41bcd596b9d30eef0",873]
+["3bb9ed1011078e34d69037012da733099bed0cc389787ff1048e8e89c9145cdd",954]
+["d9d68d484d1feed86a387d9fa7acb27a7e7c75ac2487adb97b9e97ad92c80467",873]
+["28aba7b30621fac39955229ac1346798f956de459aec5eff3e9bfe8ab24a60fd",954]
+["f33c128d292e27434360458af43e71dfa3b0edeb858f141aaf74a955c08b3fa6",873]
+["d6adcdd974601065b73e36210b470420011980691cc7f14f7e07d9a47d0835a7",954]
+["cb57d5dafba7e8b34272f19a49caa3d63f8a05c751594ab869812bbcd765cf0e",873]
+["ed19bfe608e436cd1663dc15ddf4d2a81d674c204936b0a05233f88136e1d5c0",954]
+["c5acad5320950eeeebc74c8f37273339f8963828a300929d7f5bc293316e2d31",873]
+["4f9a6f0cdcb0f5440e88718ddb0a42deb542f573e1b882fbea54c814efb374b8",954]
+["52740079ba9f1adfa1452bf03ba95a48ecd56614b2094770de8dd95323938bc0",873]
+["4f82bb424160932e751e03e052b13767c65ac0d1f3bcdb542d7cf7fb185be6ab",954]
+["01e695e34ff876a39c12474c25eed07890f7f8c8196f3862dbbcb451daa57102",873]
+["4155aad5cf40ec967921a1d0691824fa00d0cf4571ae834ddbe82aa307bae01a",954]
+["4c822dc45f845e893a2cfdd234281a852d67212356afcbdd2c9e6ffe53fdfbee",873]
+["0b833debbfcf8204b4d7c0f408e6be4d6889fa0787029bc6b230b46563eb97eb",954]
+["074913d2b406a0f3535f6f92be9f9823b7fb001cc825f7de68f6af1b9ce471a4",873]
+["9c5bb9ac00ae1714946606b7d4cd445409fd060e1411464e7847b0ce2fef83e8",2476]
+["81df337f5463e2a520aec9e4fcad4e63c89ad56433104e23a772b0ba69d7f88f",938]
+["786373522a38e887781b843929964f006e10b510e3605abf723951ccce2aebd2",954]
+["df4d65e90bbf631907d280686e6aeff4abdcbec15712a369021086edb2484de5",873]
+["ff3ddf1bd61f46d5a56cdc7c6bdc9b86e9c6251d2d791304174cc86613a37d2b",954]
+["fa1aab3433a1b00078a5848aaf11853cd160f89fdc0e84047cffcac26868b4f4",873]
+["d6eaf11f26dbb4b2b9d99e1cbe1fc70415f1dbf0e41d497f498e3b814c81a603",954]
+["9db60e1bf8f03608355400429d04b9996897c3398fcd49604bf338830b456bc1",873]
+["4d9a2359f97dd54557e1cd5e91785756cf319c5631f51db0e2e43842ce51fcc4",954]
+["6326ca0a9d30109f67c0903d89126f354c94098353198e8a9002864425365a64",873]
+["c11af8538ed8c3d23c32d17bfdec1854409dffc0c8ff61c652a782ebadf4cc11",954]
+["e5b1b92e86b866f2298c0f92615e68c5af9c96c9f3aaa0f53fb1435c9a2c397a",873]
+["271fc20ebf93be48d0332bdcea16554c39b583d4e40323366299535dd371d25a",954]
+["dc50efc1153a79699a6782f2e21126e523c01834136b4f0b7e8edf9cd7a4149b",873]
+["e7534fd7bcfe1ed2255a6e24aea90d3ae91b86496484b189d7f9ff4058aedc34",954]
+["e30869936a91e9c37f1cc6f455d57c0a06e54e86e0e17604b874577e4f1645f6",873]
+["111c03bd501aac5d79aeb6c3f8d392b402989f94fbca8faa92c5b19f85a75011",954]
+["d3ac0b21240983f758ef15c0ebc24c561c6851f17a1a22c005161cc06b9088a8",873]
+["89ac277236a8f1cfb8b91742859ab11331c270f25e17b993365c4d6e952a93e6",954]
+["edc64c277b129d84fbfce26aaee7759e934acd5d9bc2af165a9ad6d99360e4d0",873]
+["2c7ee5c8fb2e3f2951d5bfbae6cf2534232d205fe07d2b55f74ae25d7c1f8299",954]
+["874bc7f0f097fb78b0532c87ec570c7cbb8c4ae6bfdd5e1e9b01ad601a09e8bc",776]
+["913f2859d7d8ac43e3febf55748d49e4793f6e9101520e6327db5994763062e3",2444]
+["88de9c390b687b6e287fd4c3e8eb008eef6500bd1afe5952e9bc7bd88cf8fb04",827]
+["7b070ab6a2e0ab20455489cc348d328696d32f451a9dd14f276e459658590509",763]
+["a5738a223edf59bb86192473ab3e2dc55837ef53951c6f5e386312c9f59355c8",776]
+["c9c04bc91518d98cad189828fecbdaba670ef5209c45cb53eaa69f2caeeb0bd0",763]
+["0762a0357bb9847e1ef636d5382d46e4b3dc2e08403dd613a2ff873fd1f33904",776]
+["81294104636abc7699ace3e817ae3fdf0d5a7665c96d99392e7363a0557dbcd9",763]
+["40fde4f1d53084a8396f2ae566f5eff900ce7dd0d6af3c0d09d604e13750ed9d",776]
+["f3744fd3c0dc47f31cee66ef7875476e67d8268090f3d307b7d2a8050ab2d0c9",763]
+["95ddd2bd3ddbe03fab84ca1620b615e159dcdb036331416f15e897caa6af262e",776]
+["cc222dc038b014d84bc8d94dd463e78f332c40fe15680e46e0c571324829877d",763]
+["859527588c9a5234d8cca0d2fccf47c50a84426fee17cdcd829370912b65e9cf",776]
+["5a808deb4b6ef2070718509a30e0bd89aa6410ab8307101ac67d30a9275386d9",763]
+["e1143246494fa2e38dc3634e21d5f3fbb80a766d8f702edc97823dfedfacd9b3",776]
+["b1d162b21924bec647b0ff5c9b10d8047f01afc38a97bd912907cb3dbec8a142",763]
+["101f11a6907ec77cecb9e09abaf9b4aa38824e275e7cae8d6539219c020a4f04",776]
+["47fc921625d2ce225448dbe8c2ec9e691aecb017dfc09278061d44c182c730f9",763]
+["5f06b1e3d74a4866f79de908f6c91a251c49d1ad15c78f6e56d9a78a3c7bd656",776]
+["510edab90930777edc5f6252b2fa61b3d0f3613ca02434ee1156c9a6ffed349e",763]
+["2e31c87999ccc23fab56f7c3e2bf457026eacd7bcc07ee89d8be736a17b16aad",776]
+["6b7fcae7517db1d0b9c8f176d31c789bafad913ce7d9e5e5eba8f323e04c5c3f",763]
+["c06b6d76e65c2bef37177e1fde8899b5dfe4a728a164223575bccd2024ea5b5a",776]
+["db466d43067c1f959467826e5cf037357a15dc977fd809e2ad5baeeabda12b49",2444]
+["e16eb7f6cc6be94507819fffa9947e8d9664a6ed307d45eb020a2c9027c5e68d",827]
+["87487b881f5ef5fa2239916531dcd86997b26655e7b2effb465c5aa546096d86",763]
+["dd4f18a3131d05c0f0756661a7c13ed0ce388c23a178342960405eda9b4029d8",776]
+["6cf2b1f14cb7f58d909b765649f67219ecbc7c005053311b23db9c58dfe415ac",763]
+["4c9260b9d4ee644b13a643b9f030e1cb28dd9512e188c490322d08b6168396f6",776]
+["3459d1f42bd2088fb2504e4f070d5a37d8ae105bbe3734f9038b11a86696a977",763]
+["ffa0accf0ba884b05c049f310af5c8744a234298b12646e18c21617006599495",776]
+["6b3e356366d2e201540c48db18b79c035c5e32f25646d2bf394f14f54c9d27f3",763]
+["ed6ea9fae73677ffb42efe40744518f81522ed2e5efa1351fd33d4c5dd992b84",776]
+["fb7bd6b2c03b6204a5cbdf87d030b0a24b99476b3b8befcc59e7067241d944b4",763]
+["393843c8eaf3906443a246afd50be43fce262e0e570052a3466629d90db7cff4",776]
+["67e598d932644c80bae6f2201a7ad9641d72a67145cb240f2d90f20352f2ae41",763]
+["d71e0fe83427fd9c40f2b558d0effd2e3d594fd76e14da728fa266f18fbba761",776]
+["18c79be569971d4c0010a19a36642b428cff7c1dfe45780f59c7ced02c693bd4",763]
+["de1c2d90f45d176a3353a562c07327bce56b064c6b0cd02b15b48383327f5e7c",776]
+["f967a6db364ac7a82b1b1ba3863416ed49f5230a771eafd22553b6a150119e7e",763]
+["318ae50049317fb21f8572a701d597fc164a3acf4fb4c290047c814d5d6ac91d",776]
+["c3712737c7c8d41b2495626e8fa2f93f1c2a739a50ea1e8a7b3db5f901d38f8c",763]
+["e40ab6e5443fef32bfe3edb7ac1afa5839b227d5974505483eda22a8dd6aeb4e",776]
+["6d3ca2a0e0f2394b1daa81603641fce244733cc3fb26b4c80fe94fa983008c2a",763]
+["de320fb1677ddb67168b4698d321b3a784ec530248eb22c33ba036cab6223938",812]
+["435ed4aad88f97e879c5bdc9b60b5b14f4180498d46d391024710222ee7f3d82",2444]
+["c32fe5c358048be6afd42a239d38d952608715af414e7b8428779bd7e2b8a3ae",863]
+["528875a1c0d1db0bd8e6e51ea97bd91a0500445d4e1a209eb6d423c384b24506",817]
+["7a8ad5768b1b25777cc4f0f0fd948bdf878a0d748507fbdf3cfbda049ab2ca08",812]
+["67963894172f614ccc81d639bd8953565768986c36911bd10cfac2596f826065",817]
+["c3a9fd4cadf607df93a50b884d21e632aa2750f39296df60a6b7ee95fa6490db",812]
+["c70f59d9ea1e4494aa43cd3b13aa1085a0c85cbc515f7fd653f261fec6341743",817]
+["728105885505a0d772fa5844123ce51cb062711ee6bfd9dee73ccc70a3035741",812]
+["8bf011da3e93ea9c1add7f1600aed4b2fc2a7405477236b2a59384de8210f984",817]
+["3c2b25822012306d23b7b45f7626416916b7784121c1eddab5550dba14c9ec9a",812]
+["0fdb02dce00aa1631806071f83619965728deab1f007df254d5608717d05bfc0",817]
+["4bb368885ec89b67e5a0e5a1156ce18f5c7431ed32048d218940d667df72f8bf",812]
+["81a1144e388c8a9f0a5e87c4f59a66aaf9cac3c38a8976446ba1c547b99f0bcb",817]
+["79469a2b784af800661189213052e481492dadbc0b43aaac0dd811b2a71e21eb",812]
+["e2547d07871e06b5101bba2cd57b7336ac5273903e09a030cb481fd52b67a66d",817]
+["7be2f7af0fe26dca514341e2527d56ac06d0dc47e5d71b26bb4b3e28098be248",812]
+["ab22cbacc4cee15b4e6992f9d732e012142af59d340ac3db5584ff34126d0810",817]
+["6b64d936af9f84a5b10576a7b1b23f4b512566c6b6f063d2afc0b91c9d7aabfe",812]
+["48629ab3b2537dbee95c37586f3759966ccb6f442919736dedcaf20a642f3df5",817]
+["21a6591997814d2cf632fd2eee4ce26a6bc07fc6c0d7cbd854c9b26d6169a273",812]
+["e586789975330ec34954e7602b52ffe071cdf6462a3a943372b3ffc92509fb9a",817]
+["7d7231103b65bde1259510f87da516ce31cd732a95832aff4117d74447276b24",812]
+["6091070bcd39b17dc49e77b3f4e81b3afe8683c4ff0886cbcd15658cbffe9b9f",2444]
+["b42bb3624db99ddfb9f2369dff2e453f2cec8a5e43174f0c6b9658cdccbb3bd7",863]
+["f9fc122cb89ec61a0e7653408af29d788bb9f11ec2ea362e29137327d1e9da1b",817]
+["78d57a7a88415f19de13bfb45d98fecb53465fa69cae2625632007d6e183c53f",812]
+["9c5ea6f8baff4965318fa902cb5d8e867ef693982bf9b8471cab37b1e84ca08a",817]
+["5e6ae959a4b97a881231028216b631d9ed6412222a0182b20340454491993441",812]
+["4e6b1c1e16200ccc66fa51ed732093031bb74ceb8f9932862920e113dc6c8f55",817]
+["6eeb5f0bbad7dec4e6b99f989afdf91065fc333ba9466edcf4cf17038a3e61e7",812]
+["d9275c295550b0f56cb842cf0c9afc21217d39a586f1662a07208066c1ac3b53",817]
+["6ebbee4b5e412bfcbe4157b88acc33fba5606b84be244621d617eb439af16ac4",812]
+["2a444031277841d0ae11b7d4ef7a368d43ef1243a01d1cf2cfd1a2eb46ce1495",817]
+["53a74915851301438895bb9e37214c76dafeac1a8dd0cd8149b57b16bf784951",812]
+["d46e34b9a7cd0d6435412e7a6e08d07dd2f71d7650bb5cb65584f67bf1beb5c6",817]
+["93c5171a3751322874d2f4364a945f9f76f69597e26eb075af8a91fdc596deb2",812]
+["8c1576731a77fddf2b317d57ca9e560c7d961af8e88314434448955d397d7a09",817]
+["a9ebf014e02d3a809ea289f308398a0abebe53cec9a8b9d8f9917b626cceede0",812]
+["23446d76fa539a2130dc0c1e4c5c4781972fb359cdb549b613dcf5673cbc4f64",817]
+["9d396056c48e610546febcc935591fa6cc02e2d05401824f541c1d58f52fc2af",812]
+["62d93425f23ed18925adaf0be8836a459483bd269893a46f5e4a1b0e3cb000bd",817]
+["aa48b74b8c5556dae1fcd6ae9a073e8bda7a432e984b8a69bc2875511f8187de",812]
+["b365d43fff5128b7f97ca32a5c771cb0ee83ef08ca509b0dd3c243f3f2ad18ce",817]
+["e6d0416b9d13c7871e5b513afc6cb76592ad9951a5203e3f84874c2a3d318ac7",852]
+["4f72c31474f26108da9ac18cbbeb21b95fa35cc8a2366b891de5730a9e7e26c5",2448]
+["201cb1e7104f3337136e0592c2d3eab19cd6b72f691f24944ce7218ce0393c2d",903]
+["49bb73592d4997ef7a3f81c167c04f008077185fa44147f1dacdf94f9a7d75fd",905]
+["f19274a8578f6c6b955f29726b2903a9e9cfa5b27721e483532cc48a49305e6a",852]
+["6a6a7a62683f22b3c9d4bd49657cceb39c08f84cb54743328e2b6e8acb72881e",905]
+["ed03351bbefd54e8eddc403963a27ce46c07eb6a434c30e1c960b86d985074b2",852]
+["9d83c854f132844643a17b057698f485821cde169074b03aa7325c76cad75e46",905]
+["e12d1bf9286cb60b41065ef3d560f67f2f4448e093545946d4e508455e7df603",852]
+["a570830c3ab62986ca1a0b9a7f358a096a90062779cc72d5b1670888d0f332c1",905]
+["34a152a6e92041b08db31e08bee8f155ae13affcf7f070f55f07299d5d2bf95f",852]
+["da058e2389c0588e686431d8b7f814b145a4dd5f8f57458a63465bfb13f9263a",905]
+["0c44f4c5e924c187c512fd6de0930f881e3eefc3d63091813fe6d2654266fe39",852]
+["fe94bcbe0ae938fc9702a1ec673019fe6e7ff1f8cf329463b45583906f351a02",905]
+["86e05f97eee66da459fe176c1e42767c76f658d19624fd620e228038aec4942c",852]
+["c6de6095c665f9b9a6ec193777ca41c40189a464d772a2d86fce0c1db55e08f6",905]
+["400054bb9a1d9b6a0ddb34c5697564c6252f373d9d569a94acd2bf6ea21ccf10",852]
+["5deb136eb2fa235b21b66ecbc7e4fd80c8477639469c2e61f20007dbacbb7dd1",905]
+["74f52dbddf1936c5d94e719ec0acb2b88163a37eb5644904bb6f3a8c649a44d4",852]
+["cd926f5c12d0a78f7de8cb527b5da9247eb2e5a50f482e547260d403a7fc4dc0",905]
+["2aaec5b3424a41a2a9b0d68b62bb15e0bdb29907ae9cb95573fac820300c24c4",852]
+["527585b325c8b9886e9a1fe37df3cb9c2e2bc6db3364f24d80c7112d7f27d654",905]
+["20f803d0a704c48d41eaf8d58a23a363a0bf6630b04c4c9979365622655c3dba",852]
+["bb5cf283cc103decfac7570623f3dd1661078fe12628b27660add4e7a84c30a8",2448]
+["32b665c7bda761a300a65ca04ff5a26b9975c3ed269836b971ccb47633a2b906",903]
+["cfcf0abbb26f04da89d2c93568f90d4ec686cf8818727872e60572954915fdd9",905]
+["2abc7549755e5328c790fee00a51669a3b483bb23ff10db0b500396db3f00b45",852]
+["ac33c296d93ecf6cf7ef54e8bf9de6b84485a6689f0871af0e4240968af82a5b",905]
+["b1ec94eea246d423bf387551aa4082cf5595b176694d49cceb33befb5020d612",852]
+["3a0ca31352228a538da468f95999900f6b02f69966127c499eae160e5604fc2b",905]
+["f5a76398a593dccc4907d54cabb003bf0d3e269507b29744e57d5e91a3a6a0df",852]
+["a9fb3167789e721811b0e62f5b36a52b7b6f7d93431d1c01c8d8fe49079f9476",905]
+["0c5a04708402104fc7f8090bd086b3a33e5a1be32fc95a0f62b60c82b3d0d289",852]
+["56cd52add123d210e42dbf0db4e76fc0acee12d794359dca0b69e528cfd57d1e",905]
+["5b2053f7727a1163434471a06003cfd790fdd1224462fcece5246f41d8003c5c",852]
+["f9d23b6985380553799fe748bf3d31e24594dabc5dba09d27377a288a058f45f",905]
+["8b85a88c66195435d3baa66f22acc535dd7f04b707cb1ec929d608483a79ee62",852]
+["220bf451b05f4472b3151bec288e4da4cb5c4f2549ebcefdacac5ac605d606e8",905]
+["645b86b0d54afab998c10efbda15e407178803801ae4b362c431e8057183eecd",852]
+["40c59005aa6862e5ff78f1ed14af689299e52ed32aa9093538373e330ea0e310",905]
+["f00965e2faa3dc83bc94a4e2a508015fedeb2acec53474e8be1eec9d1ce00a44",852]
+["062ee9a1072a26f77c214b808c432a5f70c2a604509b2107a65ccd33255752c9",905]
+["6e19a2a9781d4f106a051f7bebb5e652a7d329fcd74ea25e764fbc9dd59ff518",852]
+["d82de6979e65810d6c6b7a6e2f54781952df7da78a72eb38db90eba06a459616",905]
+["0a92c20c924f47eaf90f72e4677931213498b3cc63a2c32d9adc3714bc8ed06c",737]
+["3d6963737cfd03b3d9891cdde4c81c829ccd828dbe2d943f2fe26f53645a9e81",2392]
+["0ae8e6fe5021ed6fe6989772025476e96c6146a5f2e439429e43df2cd93acb72",762]
+["4c7ba8ad9bad2bca06ea40417900b2bc35be611c8ee7c121ae863b8ce1807f25",670]
+["2781d9dc4990f2b4eba75658f92ae77bd9c29e9dee116aee9f72e1b3b75d6425",755]
+["12bf5e2e3dc1c253adc48b3df3d09a3808feed781c975af751b65ce2de075f4a",2420]
+["c10d673018cab3b88e465a34288d2a1d501af9142499d08217656065f0233824",792]
+["d6deb44aeafb5bb3dcce6c00e6ec743271e00dc9d052033589134d314a194e6f",712]
+["8c0e76492bf834ef91a687a3007a3775c2a4e88f6c843fa705db4b90ef72ce65",773]
+["f4385be5d151e69df6ca232bc47c39f89d3ca5eee3614c1310af2640ca34c109",2392]
+["edc9b65ee378c315bd2e7bad6474bb0a0187277534f89293cdf409e10bf8ff8c",798]
+["2fa39c7c2825d594e8dfc82745ced4c6c11ef59e3f1eda80bd4cfd2f2e303cdc",724]
+["5c40378807915b9744bb124b146e44dfa19b2a783c6247459aed37417a4dee41",791]
+["ca15aaae1a7431d89cb36472431c37381d3dd0560e77370d65e6f58cdd17de01",2420]
+["d8b28f7a53f7fd28f2f53580b2819897361c08bc68eb0bbc89b822db1d6ec234",828]
+["aa8f5412eb52f0e35743e36f1c0952c0863129727a94c8546934fc726f5a34b5",766]
+["3173d9bcff08d1e89d6aad3ed821d473578f27f447775e53266c20fd4d3a9d18",813]
+["a09ab693c4d3959cd4d8f983b2333e4344ce6e653a8e755562bbebf4ec197126",2396]
+["09515c3b539d3550f428181ebd65a605bed0f09d7698f388632a43776accda92",838]
+["7728643f1ce6880b50eb4e312b1180919d1835922ce62f157a946af6ecbfa40e",812]
+["8d324982c4f14287a8d249eb333d8fe9529cefd33dea97a525d69fe2eab66a84",831]
+["53b92f4e87490114ea13600992561b9e6eade56d286af2bf5433ffaf0ea6c7ac",2424]
+["1d3b3f4c9ccd49ed3110904a7b28d3a908e04294c2478aebf9faad13c378b6b9",868]
+["830a504da06aeb1d828b82450a0d8c5f163b29c4e779c9a42d169bf36a547d12",854]
+["1bf863c2caf9881717cadfa0103042c8335c8d48991118cbbd2d2b189b615ce4",573]
+["dd47fe8b924d251d3d0a954bc47a5df5d68c4290e9cc43c99ae559843ee80c65",592]
+["27f0c255b9d83f6dd4394911814f01910b8bce74aed65ce91a364785b54efa15",581]
+["2f079bdb7ca876f10217039fc9ee45b5aae90a938bf77c4f362b80bc8ad4d97a",617]
+["a51a3bd26ebd33c5e3d492f73f5b5fae7166854cceebae21946404b27a47a7bf",837]
+["27c27023c37c401a94b36adf78ea59f7eaca7f6b52e08e558248e5e05a4a74e0",766]
+["5d8fef372199ffe047b3a497a0aa5af01da2e81d3538135a2635c820e6b4a047",1194]
+["987030bc5ce16dc9fbfb496e4e555a1a8f93bd7b41efa3532da1b3dcca1db348",2728]
+["48c21d7f5186bec71ab6dba724aa41af265709b4652ab10445e5eabd0ff4acbf",1223]
+["3973d6c67b8738247bcd614b184c664d94be3685ef58a42abf85335f7d169b67",2622]
+["59b5122688da3c177245916e1bfa4fb232f0d8c614f1d53fc25df84f91adc9bd",1178]
+["57bc14708ab2ee4dc5e9c25fac9cda693e026199396056ee1a6fd3999ecc809a",2700]
+["c10c0aaaff4d8dbf4bd56ddb52804728e6cece91714b74a862ec56c516b198a8",1199]
+["bbc19c1d28c81c9a52a47b268010973e9f8cd821c74fc4dc74af24186b38032d",2482]
+["2527fab715eaabf6751b365929c3fcddb3b96870d8f0a008b7e0335a629a1cf3",1085]
+["3a70a6a3ec6057382015e054f757eb856b79ee9d66437e32edc7b07c73ac406e",2724]
+["517dd2f4967363edc1f2957a130c23e7d8d647f69519c518936513092b4eeed6",1114]
+["84bd83ba46d2cd2b4e28dc702ccf92c73d4e0ea00505067ef9d8ef6d6bbd6f83",2134]
+["afd4dba24650cbdcfb0e81666d76f251a4e1d51ffd00b4107ea7371b0cb74388",1069]
+["17c85761223227b0f6df8cf00cc915d4c7f94d9b772afb37dd04e90a52e41569",2692]
+["8d131103a5e637c941a7b61dbec80708d94cf813bbb8240e6acd031173e36200",1090]
+["ce9cd30815ae1fd5294dee13b3f1165dd8a443a62abc1f414971dc3072998bfd",1994]
+["f54f768e27e12e38e7a927ee81d0cb6877e1f53473744961bd31e684d9c500cc",567]
+["be8aec4d3168771808198f1627dfc55695472e21c181e0ac4f9ba0adf8d6fa4b",694]
+["def737238854ef2d169cec56a3169e9d38ceaeab66ee5c93604d87559d25a0a6",583]
+["a26403a777ae0feb80315bd2eb42c04be296b4d6a5f44bad2a7c393b7f52d550",660]
+["ef604b05014a8508cf54ba708851abbcfaa7a0a46bba1f63368d9fc260e8fd17",565]
+["e4a718708ade3f2bbeb53dab44a73df7f835489a339576417277dbb1858270cd",686]
+["9717eec638b662833951da83d4a86a27974388f4b63ced4fa4c1f8163de01d08",581]
+["808ebcd5ebd0ab4f33b1e9074f8048d940c46b1dfdcf04e8f8bd5ace60aadc47",652]
+["711f8be7045dcd78fc5e31277522d839c4b17acfca03ac5a4a712cd8908af7c8",573]
+["5ddd7ce4f8ba0534ec5d11836feb5d8965eb7bd4b01f717e50dc41d1733c4bf4",718]
+["29b57ca27f5fb6441debbe9dbd9db8627e6d34e029f2786cff1309ab552f9030",577]
+["f3f10e5cf0c5931117ad813ed2c7a19685c531d315c2a9b15763e83eb3fb8b4c",684]
+["f804117e581733643601f5481f31247a436d5886d9b9e1c6b256a22013e463d8",581]
+["5aec9b4acf7826a7cf750eb5b69f9d9e7a18e846c9267612d23aff2ad3adabfc",750]
+["9c0d395b207e490f1664f619abf64f44dbe6980c58ea101045c2320aa1645c10",585]
+["fdc8e85222b09e86f5beaf9c994df736abc038a052ad66604e205b412bcc6ca5",716]
+["831436afe5eaf17909df3a55a5bd7c6a6d9e242553fa1760a4dfab9d3127c8c9",605]
+["59df22243b7992fbb9da57202a39d456b9556c4479e71816ed02415e54aad251",856]
+["2b3ea29327fcdcfa899cf12e7bf3dff5e8a1dfabb1b7477a3fae354d0c371bf3",1464]
+["4b44993d8d67b012b40ac0ef163eb13e1feb1e9ff719a421c2b7a93244a76294",605]
+["ac16f18e5b2d2217ee04aad7b9086d12c107c98d501e8e7ccf684f18216456c4",856]
+["df5a7351535343e8df3d69905ebb7ddd61f725d0bd674a0f1b7dd3b0af591ca2",609]
+["443d94bdff4df34c390b879d37a0e7692ff387b5971f4a4620b6ef0b91ea4387",824]
+["81a7441cf3c13a85ef6247827ba55acf89b67028ce6dbec390ba131bea1011fb",609]
+["873b39b92049938e2d1376c7f990449d4eec85f11e07f46ec2b6fe2cab2fc683",824]
+["6910d1c0690b4064da788f00895b60896e6045e199018f226d09be31afe6ac9b",619]
+["121d3fb3f652da3bec2338bfcc5280814792706d268ffacac8e36069e077457b",806]
+["2a286428c00300b6dceb23672401b9a89de04b3f7b94618f8587d104faa0b62c",619]
+["fa22b48b2adfdba683af74971468921a19876547879b609c9bcd868682ab4258",806]
+["a8a2268ed2de827df750fc128dc2ea2bfa1721391fd213038817520d74e43cf6",597]
+["bf82e625abcf283228138fc00e62b2497a2bf3dd07a5c908b810697327adf129",820]
+["b7181ec5b65de30c682f3c4b805deb147431e2891ddb72a57d4aaabbe2ca5f15",597]
+["bf7088d52b32c60861cb98fbe9f13750652d070b675bd3da700346eb75ac9170",820]
+["a76154a00d049f23930bc550f09526297cf1657329427a6cc81ada633a124e8e",601]
+["e76a844f22b6d5c374bc67ae8fa4157e17551d862c5f45b1a0a6860a938e3e96",788]
+["1546f5a05d65282b8674717499fffd6c0b74964894d71ddd8353bbb29bf74588",601]
+["bfc7d97b521b478b5bd843289e111b7f7a50800bc9434c1ba2febce660ee09fa",788]
+["a0711bcb0f35c336f304360a0753074da12fe0c92161c214c3351a9b950132a7",611]
+["8106e070a8ff20b74341e8c0f9453a6eecbd625c376e1c98e0e6ee317978afbc",770]
+["5888983b5f4bf62d3e1758cce7e66005dde4f4059e12c2b821c11ffd6cda2e09",611]
+["2c40092db00c0d3abd062432183d8ef103e109a164b42022fcb96bba0528bf5b",770]
+["1aaee7adb28f655d720a63b78eba03ba24be8b1b8c60c75d16231f6518016d26",575]
+["db0a0e28c1572e2343d95748289eb441ffde84e9aaa94d0975821c28af23bcb5",726]
+["ab26161826997f610debecae6b122e6ef4318426c9f8050959279c27f85dfb86",579]
+["73e7824141cef825944a8d07b961838b54b22d5cdffafc0c0ea3f33da885bab8",692]
+["192f6fca3f95bbdfab4a6ce513931101528606b7c17aa67c3919e4ce466410f4",573]
+["92fb486ba9abc433ab611d746083d47a51d29b2144c878eecf369a5bbc8cebec",718]
+["51ff7d21c63bb037ac41fe55af5344f53e99c000a80c752437bfe7f1a6db9110",577]
+["fefebd23979531bc9cff85fde148c5929c7e8ee522e72ce4e9996f09789d04bf",684]
+["6e0ab528f8bf36834653c0c261240a283b0f6366807a36b635117fe04ef18a26",599]
+["2404b5aa82dfbd24a699b4f1df69c66a453ec919a385056ba3eab54906e5752f",829]
+["9832d6ab16ddf01c7bd72792c75dd7ac5e4c4e10d20949911639e96399721be8",599]
+["5b84438791b57eca6969ccfb3afbcaf17c650d5c0d8a37c8437d1f0714614e4e",829]
+["4bff250531b1d5c14dd5c650ddeced68666d94c30ee2c02654789dc6fa5f4a73",603]
+["205b2352bf6195b3b73818aa0a48596da03860640c2cf541bacf479905616014",797]
+["6ede14e300e7faf6082d04188c9fd57d19e3fb2c6308ccbe390eca9042cb433d",603]
+["3b29af01e443881f8e477bdbda1d92fa83a6c10683426817d9843473310d22c9",797]
+["c3defcb859dfb5c25db536b4d85618f7691839e1ed0220513184b4494d818e0c",597]
+["2f238e29402e281b446a120eaf7189307c71d3cbbd5b2ded81844aa31487d076",820]
+["54b723629e425afb4b3d50a6cc5b937f59237a95ac04ed18c7155ffdb0918beb",597]
+["f766120145fdbdff8f5f14a78c2a8f00f5af4a4d20536dfd90a5b676a86a67e7",820]
+["a0e9d907cd74ad9c5d16a2de1c576c977deeea74cdd0f2999ee1a5b1a29a4cf8",601]
+["8727967ee6f58b1b2f1af564c518062b4bb53bfa684a7e1a9b12b5c2e0a79bb1",788]
+["113db285c528cd5cc750bd6b6a140f6730c5dd4553a652d1423682e567afd3ee",601]
+["864c21b1fc35685a371fbfcde98a7d01afb9c9feae98bd8acea4ad3aa3669b66",788]
+["a85853901f3bc3fecbac51f9df1117b2134d2a87f5a45ccbb6145ea796c629bd",611]
+["5dfe956eae7d94b57f439de56497e372a94f538df3fc463d8ab5ac6ea9d1e776",770]
+["628a3a930d5eda1a1ef44f91e66cbdd731a78deabef6fda6aeafcf26d3e63e7b",611]
+["b872375ab33585d26a1e10bcf8a274634ffce43c9a4b77ff6ca5d29bb25f49e0",770]
+["a648d1e070318ac817be9c18ee64d7991436c6e5cf8a09bcb779d06037a1c464",591]
+["8d5d0d27786cd1453843d71929ee823f7f864cf2bb31388b3f35924e5a78239b",692]
+["cd19bbabe2d00ffda1c7f77f3cee3e2afbc7b09cd12918470141e029bab82453",587]
+["34e11547bce5d21764f671e36d47c30191aa492a57f253068f1a6f9ccb47d4cc",676]
+["bfa3b4b7ebabd1e791bf87ab0589b0fea733c93acb1c5c750ee715057f8f5d09",581]
+["3a29f56af95dac7f78fa65c80cda474d1b2f71846240c18e28d1b227bed13bbe",652]
+["cecb0a54b35fcc2c52dda11f0241c41639350433f9a507ed4fc55498a04d45c3",581]
+["af376fac111934ef1cc375e57377b0eef4fd31e226db9eed52811fe89f924953",652]
+["2f437b0f9dbfe54b99786fa98715699b7b45943e1d9351b5c154242ae9b68811",591]
+["c9621fd5a5c19b142c608537abdfdf2f1186bc870e2714c52a901f1549e3149a",793]
+["5c14335f2368bd6fa836b8770a5451af1a537ff032573647828c989a7dc2e667",591]
+["ea30b30fa5a060e7b6383d745f658317ca131507ea18f7f8e25f0c46fc2ed1bb",793]
+["2fdd4be05dfbc5351d3b6eb532488ffcce0b5dbb698e31e986ac63b2c5dc93e1",595]
+["5666abd32c6edf589f608984adc148152e24249a72c450c58432267623198d64",761]
+["4eda22c293d69f177655d0a8287858238e5902d2da10d45b60ee119c3100e068",595]
+["f1031788d332933e1d10c90674c80eb65d82624dbf6b934c3855a37adcbdcf4d",761]
+["8690eb248ee71c3d9a58a4081022ce56f96d41de30b4f1ccd23795c12b2c6dc7",589]
+["3dd5c2317b74874f3bf2e2473f8e2f39059867cdc1d48583e46d465c0c868066",784]
+["2adfa2f6070ef9b450281008b368ef126cfbebeaca201fe5e7be8da8f01e99a7",589]
+["b5d07b7f02e85dfa21c3fec0762bf3071eb5cba12809d3abb9bc6682539e7151",784]
+["3f66d3bb9c38022bb85a91f353f850807fde8f1e1b0dce6b72e554cd181ceb92",593]
+["e41d6405d39a6033ca25cd67f64b493e6617e98870177761b9bef4261475a40a",752]
+["baa29675ffb38c57e171e6e19743bfba0395cd6e5cf44b354c235bbca222c322",593]
+["04bfa673a9761f323f311aa10d7298ff9604e5740df7c0e6519b621c397c7c95",752]
+["acf59924222536b22a888ee24ba8ff17e6a2e73be29edd7feda7e6f786dbc0ab",603]
+["497cccdbdd24d92c4fbdc9815bc9acf1aece03211bd4fd566202e1ae4207635a",734]
+["36fecba471605481bf90d6437e15dbaa014069dfec33840d17a9cfb949d1f720",603]
+["d1e5be271510512c5ef556f7342e11e2877af2f875df9f3e94848b03c3d21ee0",734]
+["bd0f6dc7842b3389ffc1902db10600443438ede5921f82955540bbb2d1864d88",719]
+["9da6900006a0b87fbea5c3e88af421a4ad7bad626188392d982ece7c3890e5d3",2348]
+["5bb5b2fbcc3cd6d1a274400c8bbe0d4390c80a2d422cbb5134314099c7b5caf1",669]
+["3b169a36d4b476d7e85b7328cc067c8912c38ef906e3c87bca6bc70522a39c51",610]
+["0e93b4b3686f04e637a73098f60b8259f6b48bff310fb1e628f1f0a6852ed858",657]
+["86f94069f9a427385e4179120afc886a61a1818ebbb0ee2d7090b55e14b5dcd3",619]
+["d4d1c1968ad6d72db4c998f2394d454984a3edfa89f9f531d49f631b3626e173",797]
+["e0038b47dcd506ab5dc0d8bc822a20bf126c01e595d4548d407d3dded6559c31",875]
+["4226de4794196e8b36ed16a3e29f351622bb2244b33d2f5c182ebdb98daf5018",809]
+["d6718b4d9ab30d02865faa9fd8eac6c95fcf0c30e87bbe8431686b0ae97bfa79",943]
+["43a202b57cfe0fa766e1919be4feaa4618df62d0b138fadc33a70886cc511d23",782]
+["fe9b225cb32e1dec62c4cfd662f60f790ab18650a74a28973e279638a5dab3ef",2408]
+["c6ca13f55af58219b1c141f97bb3662d284e39196e6f2334cd2b73e6ec9cea5b",756]
+["974eb38333b8a8f140627121d8d2ca9e6478bec367e9de3c226f3a6a560abba9",836]
+["107104020216a01768febf219a51952c93ad8241051a5349a16fea5d443ff9f4",748]
+["0066163469454de3c72e88678110e735626084c73817ee3160f01bd3132ad383",2348]
+["bb312f23a19f39afab702ae0c8d3d5b0c321152e87a2d4e7abaf8d576556a8f1",694]
+["da8982670e0b48a87f269242600d3c48a9adcb12a15bae0f645ea5a6214b9029",644]
+["5ea747d3be2473ecb7a25230a7ceb6eb53b7f70ee3f469cd0912de0e3854e866",704]
+["87f3750d25adf5f91a2bd22c3346e28603433fa3a59d0888f84ca204a65267b5",2352]
+["e6e825f425393d827452d02e1aa40196aa6af9ffb4c78691d909236a893fc845",721]
+["8886071cb0307a6fb540e81858a43912cbd74c2ebc2608c389c2bb9d647001e5",533]
+["708c5eb7cfe98d5f5f6e0a11134abcc0b044f1251bf947b1bfe29a5b92cbe442",855]
+["056216ed2e9a76141c62806fbf15fdfa8347c39f52a7fa2819bd3ec50d2f844c",840]
+["bbae9f479709172e48e0880f3a34929c0d205811d73829bdb7221794b64c62a9",853]
+["2924498767b249b88d46b09a80734c2c83bb8ed27c19e15aa5e4f6f92976f509",831]
+["6b8736cbe4e2f1d088d4fde818e2bfcab4d80558c549f62d99924ea7135b89d8",874]
+["ef3dba4f218c967bbede64bfe569f487eff2a08da16f4f38f37fc763076da021",897]
+["425a4314974c69d33c1c4f81a4eb0b9b76c84bcf0229cdfbf6bee4ff50323540",898]
+["161896e2aac4999b10c648b9a7b6ed1e03f02359986a9e2dab6a565997c2c87e",977]
+["1ae8e6298f732db351d4d3a244c6f1f59f2f5737c5617c347d7413a4bdaddc91",567]
+["87580573c42543785cbe0e1b3c9bd0cc4c32769c50f883a6ca15ae28a9dad435",694]
+["651a2a5f9421b3148c129ac3135bb092eeeeb837231265dc9410aadbad25994e",583]
+["1ce7f78f852770b787eea1264996611b7c120af8c8a6914eb278cc351b16b81d",660]
+["8573eb86e39b0b06143ed2e78b89a9124b52fa6e092e5c865493801f2f5cb102",565]
+["f2d4460ce8e24f415505bddff2f98845f00d70412bcb1618e29aba91a6de9a62",686]
+["a82df4755727342870662e247252c680f9cdaf6194bff1820d662f8477521292",581]
+["22eec870a9df1a93c8944cbc96c0acb6613b0c5a0cd01fcff9558706a228ed1a",652]
+["0667d84c2172e556b1d4ea96727276bf8ebff9516b1cc4cfee43470d8487a1a6",573]
+["a0e93c2cb843eafd22d1fe080263e09bddffe4d4c28238838b3ac9c09de22702",718]
+["c23c833fb90e387949452a8fa3991f8f900ff1c2ff47b20d7bd660d16d13ed58",577]
+["50b03a8df4322fe8d1051741775e2d8d83dfd1e2c568b76408922fe2eeb8ae38",684]
+["f61bee7c48b141a9ad7deabb16c77b19cc7be6206e0fd6feafd3bc9815719d7b",581]
+["e2b4207ab0a1ff204bfd8d4daf3b499029668910f1c5d5e541101811fd529a2f",750]
+["b833ad2f335352fd09f13693403db285407bac4e414d287ea56c81abaf7b7df6",585]
+["96eefb30f2ab8e4a59abfd83be9b1f2d0befc69bb9e34803435870b85c237b23",716]
+["13ba0758cd1ead4f1b372e04745be315f76caec036dbee16e04cc45e350e6987",605]
+["00fa1370633403e346b323ba486b917dfcf540d5f65e5ffcc831bdfd07ce20bc",856]
+["5c066676a964a8d80b780088c249218c20235658e0caf6064f67401739a464c2",605]
+["cdfa3d8b1a033cb74f14ab4c3d67ec03d9b4eee6e8f60e54440038b1776824ef",856]
+["cc00d3c93be9271bef3fa9a8d4543fb3feb1fa526392bd19ba9c6a036c556888",609]
+["298b6dd61776f449d634c1d75801db85185f9b56682930c6905d69fb673535dd",824]
+["04e775caaa1ca9eee47ac5ced1df0d079ec3503b7ea0d0aca32dfed5ca94012a",609]
+["481b686ead934a906fa2248a7f8ef7151b8e1c0b730ebed0324b919e3dd6844d",824]
+["d3c6e07268e07e5bdf005925b254a3837d835e9a4e1518be648554d34f539981",619]
+["a1d1cd2d2da9f6466989d0085d117b9fcbf9905ce99df37f441f55cd1cc07527",806]
+["ad45e087a6f3a30dfba1a67b709ce46935edff616926982e1b59136ea5f1c0d6",619]
+["59650259fb78216997d8359f0c69bd65cfaf2001442d66696fefcbd46ff9838d",806]
+["e7ffcf1ffdecb2bb5e1fb343abfc9d55708a20ebf659688da8f8bda2fcf9834f",597]
+["e2bef904a37b1403416a759c6419e211c4a1a768c3631a2fbacd72f99dc1a638",820]
+["3280f57168715b794f4748d4b4b6b2e70ebf878f436c1afb896fa030a411939b",597]
+["a596bd5313a91ac06a2d491500613f728ec6761d0b641f7a849e2a4f9cf60f9b",820]
+["3b65d892827e3a231ac72d20e3ec5efffb5dc9979639637cd0ebd14ad6d0a65f",601]
+["319667c3f0e57f97eb5058ae9f9c68952671b6be1e35c285c33822ed3d8b7d3e",788]
+["e79ae34ef71d569a3eb932d47140ff92dbd14817c9707865797486ec2ef0c64a",601]
+["1c63cda459fc87d1fba6e279f65dd26a99fbd2afe7657e0d96162f066a2f6bb8",788]
+["7ec86927281f1986a250e88f4552ca7292e154baa442effae16d0770b2ab2459",611]
+["6eb222f9e86dde06deb1c7438a4938ec56199ffcd9d0583fba2265ce5ad1147c",770]
+["93d58a92ec2778eaa758097c801f6b286ae096441fb5afb12a9041da71e8ab28",611]
+["f3b79187c7bb39733c412d6d9031a9f8c897073575a312f440e38da2ae5c4bb6",770]
+["30bb234fb43ccd5d01ba4eafb7d14d7409ab5ac39d6b8ca8d3b523eae436e498",575]
+["61d082f4a5d5de12e40f2e9ecb4f91ce7c813ea08aedf2f35cdd7698741f0433",726]
+["f3891bcd566a7ec75f14957555506f1c1afc3c78c1f8aa0cc9d8e0cc960e88a3",579]
+["b2667efb4a7dc1a01693dff541e8c70cb9ee9250c93dd7d698deacc26a1c45c6",692]
+["ee766958385cf06037364ac0b2a4bd614e57ed5bc0507aad78f9001650a51546",573]
+["2b2ce7e8cfc4ef9a64e67adce513363ebd16b5210c423653fdbdc018997bd45a",718]
+["f5c689a5c5b97deab8ba55b44e2db70862d08090b7584533cfaa543f9423bf67",577]
+["70665ff8577185660a6fab371acc1e5a711779355aa75f40ed66ff63b17fe0c9",684]
+["dbc20bb4aab2abcafd85f4878769e744c456b253549d822710f839070ac40af6",599]
+["4e18d3962a5154f71a0c0132b962d32167ccb31eba65b14b384d03b0e0a569db",829]
+["1d6c32117c0d8056157232cf454e05db029fc9b74076f61809fe230735fa941f",599]
+["9844afcc2400d430db5970d9c253bec72c30b5becf9c54e73902109be72064b1",829]
+["dd8a6b076b69c874fec2b40fd17f9e5fc969efd4aa4727ea25650523d2afa8a7",603]
+["78ba42ce2cb148e95c31b45a8f387623dd9acabe06306d47be52012035f187ee",797]
+["95645a32dd02c1de191ac6c952808014b53ff5ddb7e79f20db9bb1e6cb907be5",603]
+["002ce639f6e8645c27812bc8ae3b5c43ec4ebc768afa817efb04f8a2d7e234a7",797]
+["952996eeb5be5bae583153de73ea02df49ba964a589cdd461e0960244411333c",597]
+["4aca08f9253d03eaac2175133c0dd1abe15917377e706a0f926f7860eb693b0b",820]
+["57b913f45bc55e1b6ee0200376357507e35de8a15285ad3576ef53d9cdc78e02",597]
+["83465c10d81a721a58dce0f9899f49101269a44432b3b11b55192ac2e4206e5f",820]
+["5d39ce09cd78f377be4eea748d217ad1d7db4e722bc4b6b119ae26133efd48eb",601]
+["ec22c2ed8d3d782aa37dca913eb243b5f15d22150d39f4436524371cf913e9bc",788]
+["29e08a79858ace859533ca81903201ab9ebd4b9656a8035cb876e315725aef8c",601]
+["c1b7d1dfc1617dc6eddeda5a94dedc706df5f830ccfbbb7e024f83d1fdd3d35e",788]
+["747893590fe57bc6a48486db33241a5c7676cfd14f0258b4acd45915b1ba21e3",611]
+["f5566b0346a3e34a3ba75946ff565a0bc00dc595ba62a1c63076b8caecb7cfe3",770]
+["73fa8e25c6e3bf30ab45b50bf043d81060c5298dcfb3a91b08605d7dfb615d36",611]
+["438339854f97cc1e6b63ddf180d625e335765b0678a17c275972337d9f1ea3c2",770]
+["c0e54e6f8f4d6c915971e37d604ea937767771f50e519a05027cd223b705697c",591]
+["c10411ef8355e759ea266343683a7f30e971822dcdf3f47eb91c74bc44a92c95",692]
+["1fc5bee86267c778b509a162d147b3d178e331a48e773ba70b67ec29bde86eb6",587]
+["e835563e9676e88090632ce5c529840da1603ff6e65b3b88f2d722c626bd30d8",676]
+["c878da8ef09cab9912cf4ae38ddc856a50696a4c00e1bf67768ea945ae29edad",581]
+["2916685ee0644fd8029821bab02c8cb4155b0ed1f3cce20b10c9dbe58edacf0a",652]
+["7e288a92a4359d67fe1ad70505830f699f3afa7f6e20819136b35a3282676261",581]
+["1d08c966ac1bea96a509823b5dc389325f2ac8a059d04619a8ff3e428e4f5d45",652]
+["5f138af95ce5d072abf275673c07348ad179379636b291f85a4a67cb09a59bd6",591]
+["c5694750d651e07a1387446ede1e3c893e9172a351508a6232c31f609f335060",793]
+["f0c1846128256fcc431ce8e87f1a9da5638f9f405b473daf48961f1b7a09261c",591]
+["ebaad10e1cabd1a3e7fe56993b9ccadcc2a45b2b5d815b9dad96d9468e4b473d",793]
+["3908a457362c0c286fadb0b5b1a8f2184d810e06ce51ddc7c06ce2823075d4e6",595]
+["d477e1b797e3299fdebe7aae0f19a0386ed14dee30c3ca6f6a077f56f75b1f51",761]
+["a7ccc38e1f2b0a86f082d17148ef13bed5b1d9bebc5c4ef644f49b40572db01f",595]
+["aaaa1b2661b20c837fd35c88fe70e0ddef7d0cf43c3e3bf5735b68a7666b5e37",761]
+["0cd9f62b9452bfdf3e516f18a347815e5c9d6ee0a256501c2bef23f67fa94e0f",589]
+["05d5eed3d7c352bf34e87c09a620a352cbd3415b90249b6946028652459909db",784]
+["2d7533cf9f062d058b43d44b2db826ff411e240cbac380f76dfda04906ff1c75",589]
+["279254e631d7694d43d6e1e251c21e99866efde796c43a4e266afff5acffa00e",784]
+["39950e4a237d8a9977b99849315b2f7a38c0ff37c98349f4911a02d42932c933",593]
+["c0b15bc42ff1d452d452b1a7818ea9cabf1883061021ee0125fe4c27878be9ec",752]
+["f83f3b9b4f99dd97b48d928db12183e5ad2d7b11ba74917e9c8686440a23a583",593]
+["60b24d8c8c57562a33767757ed123e9e99a4f21ee3f75b76b666c9438f479f3a",752]
+["796635c2ba3f132bf0e062e3e68adf33cad7d34332718eeea82f02deb4366f9e",603]
+["19d7c6e50b854bbeb53ac2025855e54a45d4403e341583f44dbd6f97bff78650",734]
+["6c5e4b666172ad1dde737ceac72d2e7f5e2dc54e1db438826ec973d9bba1cf0b",603]
+["a50333c7fe7156404b837222dbb9f50af3d7dd37fa3dcd362954293d55956852",734]
+["09af1c611c86905741777d99b0fb86f0e00bcf480da5ba1154319103f6ac783b",897]
+["983118fa6230d6218d25f02366e9721aa76aea02bc5987367881ae5ef30588e3",1657]
+["d86c5dd9c613e112ae36206248a4419c36b00cec1367d6a81c9393ad97ce5d97",897]
+["3a965ceee51d4ec4a4efa3cdd9005986d80f2f19e8624d138d3175c117e904d7",1657]
+["20f5f523f63bc1b68655ad2d9992bf02b20bb375b74dbd3351366f7270faad98",859]
+["34cfb8c65e3f6b561abd275b65a724ede6dd0a7c6206e950d4ff0f38f77cecd0",1509]
+["291c73c8fdafeacb68163505700aa0ffe8bb696c7fda23c443562a10ca91d936",998]
+["e6ec391ff52958c7322a0f422585a34c1c3e3d512ae70f731284d867362a09ba",1855]
+["37fb37603e1de4f729e0fafc03e70926c5429b45605a75c1b6f4e21147dfd0af",995]
+["c2c0eb9093b06514250277784de1de588c10d75b80ee1261929108967f88acd7",1829]
+["f00e17e7f049b87697ebbfd25a6da2f5eddcff828915465beeb9bd327e22d661",799]
+["ce68c2f2b94b6ae3f317ca1070be549ee0b2fcf40f3afed6a298438ba6f18721",1312]
+["c1b5f4c94d6ae3f488e7157c4df34bc714cfd6d2cff07fea2ce1a3b55c574b7a",799]
+["8c365e77356970d57ed7af35f8c0b311b9c4c3f31c4100d05f9e492020c953af",1312]
+["e6c0db1bad5e2c0e965b167ebfdeed4aab03024ffded704977b6c2870717bf92",799]
+["5a6fd619b4a9b9f07436916bfca9e4274c88bc59b29156e87ccbd49fd085e0e2",1312]
+["0a52989fe6f4801798edbd73f498b3277dde9fc241cd227a5bf2105c0d12e963",799]
+["47bc950a58a4a363a82d92192b932c087831a5577c362c10076f0a1935cd306f",1312]
+["c7882e763c8a8b5a20cfad9505904c25c8923be145ee7901a05f69df057320f0",932]
+["8d187318f827b4597f27701311eec6c61157ef560b908f97c0e7ff0cac80668d",1669]
+["14be7743ddebb84d636338559805a50dba3051a9917db08ee9db3a2d2cfc4fcc",932]
+["e431f0a28ecc3bc27a1c6d4e7210aeb26227bf33beebed8bc3b4c624f4afa081",1669]
+["f7a855675b63ed8f604196171bd2fc64105143173c8f8f559feb9a650787d55c",932]
+["ea98ca055a7733be0683ad46807268be1e6f0a7ed09027941c269dc22a7e2190",1669]
+["4dd51f2637369bfd2c2985ec272b16fcea72a95ef7df6c7465c34610099f7c22",932]
+["bfd4f10642ad3d7938a9a750eccd519aa318646df14c6a4452fbb73890e917c8",1669]
+["061cf4787743264c58fd0bfcdca9b632a114c13d111ce258ac3383da25cf0f9b",877]
+["cfd3048877cf836304b7462a8eac32dd65ec4bd8c66477ba622c38d8efec9a98",1467]
+["13127a7671bec4f9f211600e4b1655a38ecce5403debdb2730f71146ec544b4d",897]
+["f8a93ad6143cd48604d807a9699184ce8da8407b5353f292410d7b4057d13088",1657]
+["e65bd3cf09405eefcd5a29b9688b6e96ca81c589a52b1a18c6b235f79c0bbfc4",880]
+["dce7c6a87b64cd50175e852e89852370ddf6305fcc46c7eec7cc6ed24ee5b84e",1521]
+["8daa364c95adf3e6b252a03e638a3bb0e79f6b060ceebb98a033ca95aaf99986",900]
+["a94ead2bfd574c09f98bf02c3f56de1fb60f7468ae65bc46dafd599c59437899",1711]
+["b168d65cdb1da105a8d678781db979f3ae91bddd42fc4c0859b3345a9ce1799c",877]
+["a747e1b0f8f9824efb03ee52921ede37236b499de9c4543a715f3cf04dde587e",1467]
+["314f9abb15ef8daac6940b7d05421da982d142acf8dac1c772c01dbfca2cfde7",880]
+["755c29c3b0b5bb8f6d64baeb64b072434a92b6b6fe4c44a24877711ec2152706",1521]
+["1067fb0f3df611effec1cba405364b86c8bb348f5c59ab0e27681c6860868f78",900]
+["c7e1c3e03c104d782b3ac1e6097a81b739d32cbb800cfbabbc8cad3c54c8ead1",1711]
+["ccda95a659ada5e9d268cf97bb195dcb3ca22d56acd2498f978bbae80069b52c",897]
+["d5f1696a7e300df458e5fcb52114b18ab732b094f5246c139c9b305eb3ea32e3",1657]
+["cc0e2ce666e56822e503825da0e0d944a8613cfdddc4d0163305e55ee7fbaafb",972]
+["91e1b7173fe859a9c638fe86f6e4407a105181828b62fb129ca139119ded23d1",2053]
+["236b5888ccfc7c6af44ffd1bba79500959d41689d11cf52fbd8a2aa89674d694",972]
+["a8575cd985fd33691a90d0007c5ddc63736db46ee7e34852f9fdd61d8bb488bf",2053]
+["c097328d19ce77da9b0b126f8669f595de79fd8435c64af3f74c8e21dd12bb85",880]
+["60e14a6cbc40dc4baf587c428b940ffdf02dd669dac59775f3cbf41371a1d282",1646]
+["6d475eddafca58a6ded3e5c899b344dae43daec6a3b819cceca654c0be7f4b1f",880]
+["4b6f058cf8c197fd56da29806430fdb2d48b572c51ba7fe73b78f59a27baa8b9",1646]
+["5c552281549bc0501faa2b17f5815c43fd6160e09fb37fc70fa3d659a5108792",880]
+["5b10b38bc87686bf2cd23b462a81a6101ee9a15fa5e11de0070d075231c12b94",1646]
+["c502c564154323e8e5d4371213aec99581c3c06a6cf5f070ec4b1d9181446980",880]
+["143a5e7f1db9145a1c9ce98c9922822f3efb3776bd3d42b27d2e1597b7b0da10",1646]
+["535cee3f87fe2462a039d621cea345e60650852410c605e9afc76a498ea9502b",880]
+["782169cde4035daff256a550ba43e1fa3ac3e2c3e9ce961fd001f0b963f68578",1646]
+["109470091f35c6ea93de524b61bdf2195ecdbcf28277d48b236ccb7f37e80223",880]
+["fe2782100122517d2202762a002679637f48aa8e71359ed6d80b256e8209a866",1646]
+["f0c2834b6acc64d1dddcc63817bcb1db9ea5045569d0684c8d6643d1e5402253",880]
+["3a370a2cc30468512b0ae2ef94a8a9c9ebaaa2faf7f1df0d8ba1e54cc5fdeeb8",1646]
+["8925a0c3bb341a122e3a4fbf37070efca00060766dd972b426b398fa85aeef00",880]
+["6ad3293d7cdd7ac935248f5de26ff1515298def8efc9090dfb324ace88acd443",1646]
+["80b2e262efa13343efcbda007fc1c84cd06ad27bafe059d9cb526d361f54568a",880]
+["ad1927177c50708e60e3ba89ad56782af5d5690f8e03456403ca324d204ec0e9",1646]
+["22f82c55b59bf25abc129593b2b570ae0bc1b12327a8787a8e339723567d92b7",880]
+["d8b5aced1dfb7d1581a5f977fa2676a956ca57159a95b256905af7f08d9b67ba",1646]
+["e9c00963ec08dcb44edbd398a7be8e7231d89612e2d738c698dae1e01198d8ca",880]
+["8f330e011ad93ae69e558abe168522ebb0c85ac677a02b2b2cba84d49d74f699",1646]
+["620bde3f26e09dbdedab98b1d50184442d7c41ef1fe5407b79dc2c6f6e9442e5",880]
+["d06b675ea1e4ec1971a9ac096a0e369f975814510ac724bf81a30e6416d3827c",1646]
+["2b14bf35c689bb4b68520a118b50dd1e0287b0680c97762d814ee3d699dc3467",880]
+["582d1a9ae36d48cfb4a0eb19a6691d03f41c39e7155e666ad762d6942026f27a",1646]
+["d2eae92a06be02e60d1a84c393d14df4e2051c93d24eae7dfa0d8d0b8e13d6bc",880]
+["cf8f9a44639a7424b6e8bc059eaf4d8278d49b977ff4ea347deb3e8e2df6a19e",1646]
+["f0fce524f317bc962362e6f75f9a48a789b95fb13553c50467aed124010228f8",880]
+["80e46987468d40769403096fa6a2680e3e87b628a0aa36ab9fdf3de3f238e890",1646]
+["732d003b6c17ec91d6aa03079445007451e4866ac5b275b37354ba7879733db9",880]
+["f87187477d676e5623b323402196f9fe560b99a7bce847669fe7fa521bf888c9",1646]
+["a862b62529477268796f2e552df9ee185bcef29a76ed8f2304c904db4d391799",880]
+["4fbfe3c4f175f4c71d58ace520944d59c4ea1c622ed4d795f60fbfc2b58a47b0",1646]
+["e297d01c42d51cdd09e47e986e89b25c50e4a1ef4adf9294401f3a6716c28c2d",880]
+["88e03021bedbd161e9fe6d2641ae076a4fae77a96eefb0436b78fa4536bc172b",1646]
+["59cfce61307d9292a1842acd731d47f7d7c57b08fbaf010bb481bf935190095e",880]
+["46431778be27c9ad3e209ebbdc3dcc4e7c7c11d60d8f4e392f5955ee61fd90b1",1646]
+["8286b192e775b44749ce5b324990a009299a4c1f3a608794dac4794af7626abf",880]
+["13324fbff050a4190e6dad740a95f86d0d619dc78ad4efa148e216fe3f9a738f",1646]
+["8cb846c204c08f6a3a003cb249cad030fe7eab1be785a60921e5575e0dc2c957",880]
+["3dd9e70917a90605366a9f97c65ed9d443d3757d0af86bfa69c78ee427e9c31c",1646]
+["ff6f73f257de1042d0eff090066b257c1dfc072d8e8de6f2ff86bb9103455a06",880]
+["4f6c21508fc477bc311bfc93398aab1745bdf11f0acf1557893b9157d58f4017",1646]
+["097b755be2b015a346d1721c11721d66cba23ed455380d03a7944a9b83b27f8d",880]
+["328798508a65c61cf551b7dafca0b427d90598bf7ce32726b0e07389594e584d",1646]
+["45e06f1cf3a01a64668d9c56c5141a3b8772854cef3ba1f3fd26fbe5f5c7050a",880]
+["ae35557e2203c8afa6f1b38fc2504aff94fa3e197ddd8f1d13327c62d2c4892a",1646]
+["60b4bf380c1fe252e919d9d971b7960e61b41a773afc84c5fc6b560c5a341b58",880]
+["43320a495835fac119e349a188720aef576b8ab18731f38278d06304eff752b8",1646]
+["339986f86107250af4d82e3b703a63789714638dabb1988e3a5f222a5ce8ce15",880]
+["d31f1d5544cb85feae43e6eb46e792c5e812de75102b74a1e40f5578cd0ca62f",1646]
+["b7dd225cfdb7f21aeacce27717f03f846a4cc7f246a59abd7b97552c4aef2cf2",880]
+["964bce186b34243126b7fadb8841e99bf961e180845a6f8b75e8e717b34fc921",1646]
+["04c692f66b7f4d4f130c7c439920d2880e03e72ba8aefc82e24b40cca510d2f7",880]
+["c3b3471d15bc3a8733f01bc93228684a9285bca31bb36cd1ee930f637ef12d3d",1646]
+["770ac38bd46886fae6ab59766a3d0dde6333fbdcb519e13c5d4774fb09d65584",880]
+["a41921cb3646a96f9499f43699a486fcb2db475f6bdfcfdf55339ffb9e4afb53",1646]
+["6aad0b21a022ae9f8243c093138f402bad1446f5e90a00ac7cb19d5019a99d7e",880]
+["a18bb09e5aa60e9cc660a12200e42acbae5730c0f8ecb0360f339225890c9fe2",1646]
+["154f511cde2f55eb25e11e542e481420f1a492d0f82e08aaf6443dc291639dba",880]
+["abe5021aca40cf995529af41f4a0c700cdcf8be150c99d15789d8072b4c5c386",1646]
+["042dfeb36c13d111a97fc75c0bce905eed6bdffc75807e70142e3bc963615141",880]
+["48e3f06c8dbe6aa8b3cfc1c61287c8f6911d8defeb8bad7c854958a1df5439f8",1646]
+["4a0b0ac81fb307734b824d11bebb8dc0d247c281c5c8421fc7f47778ae4f787a",880]
+["5cea9a80c73c11e03aebc7424f6e33f5637790115d2936ede34f04b173ae0f14",1646]
+["00872e0a1a2ac9f9ff193daa37847fe96cb3507a9fd11a6d36e2e5e126545eff",880]
+["ab516fcad66feb6dfdf9dea73422c558a8de560d3aca29f17292b391bc6f26c6",1646]
+["1579500306ea8ac32e17023a2c1bcc6757daf9409c611f373a24c85a29fbe23e",880]
+["cc17033ce99f92c8e4fadfb983d2717bcad7b8348568a18f3298a7ac4a101e68",1646]
+["69997d85d5ca343dba32c2f0369143cfdd905e519aabf74fa893efb8df425929",880]
+["21f1ec05db704ef58d51c8e2226b351b23f8f98f74533264be950686f190d35a",1646]
+["c32b0ab34685afffc62449c814a65367be2a3f607af20a4e09d3ef9f238e9cda",880]
+["8f531f3282bc933f25fa74bf11e38fbf05ee52fc5d48de7d5a270565be291cbc",1646]
+["3241e7d123b09e279ad718174f909d7c660f830c3ec14d69dcac787cbcd56321",880]
+["da88a5b2760483168d4b6f2f81f843b20f39e72475284f70b9458b5c30a841ae",1646]
+["1f3ed7ce3bc1c36a61d748c29906b862ac6fa3ca1eb4535d54c3eb3b07fa26ce",880]
+["190e8dde34bede9b1f17780d963d7ebe34c45765f5e8cda216493be97ca4fc26",1646]
+["503bdc759c64a2f87bc14be2e7c9c5b4ef244eddb3ebd62851a48a465a642ada",880]
+["4f4dc667b81c517421d3865aee3fa744f510b7bf831c15d6cdba475cfd80ffac",1646]
+["68cd9294e82ff2ecb6a3253928e04786253b6c6a5c1656d27698bf89fc9b7ed1",880]
+["16401458c62492b86c5e5eb98444c4fab8b0c809e818f176454982d707a6608c",1646]
+["c9cd4660a925ced0d400cc5ffbca63816c7873060e8e456539c1cabc5c2412d3",880]
+["2073d810aab1d05848b6359321171762726e964f42e9aa3540319c9bad7c9e42",1646]
+["f73ad7ec8bf4133c3f873ba126b1fff70468072378c3e86c8428de52cc020157",880]
+["18a02bef17ceb419f5e274359380257481a9a26210dbbc1f2b3acfc36ab3c160",1646]
+["0e7b59c4fb83347b520679075fa45c3b037db7b199a4fc928ee3f405f2609289",880]
+["7816a86adafd2fca3227b5faf6ae1fc06ed019f43f944613de585c57cab09a6f",1646]
+["105c97d7ff2ce01b6eea0834b901af6afe96687e51aa61e49baa5c782ca11735",880]
+["a8a61e508ab32b0645c4d1a72d50b8c40a16fcbb6f1c6393e2e466a8a39c2239",1646]
+["3e52cbe7487e8d20bf8a9cbf654edb5f4ee89fdfaa4b433aa5f174dc3c0df52a",880]
+["17880edd689b273305a81569cb079cef5efede3e04b32ade0529dbe4a11b8565",1646]
+["e9d80626bb5431e4c238261e2d80b6629deec456390c9aee7ef5c12bb3007ef1",880]
+["f9a5bac06a12210bd80e98278dedd1cf9a1168d7d34b6c5bc81f9cb61b173440",1646]
+["ef86f316dfa050179e21d18ebcdb03a519be9de1bd05dc162671722faa992ea3",880]
+["edc51105887b65862e4fc1c8349e7899d59c204f0181777dec84253629275b81",1646]
+["c7bf47b1caba93298a25f622580e3685a3a8ee2cd5cb8cf0a0886cf1096acb6d",880]
+["4e8f45d3d9e9d4a2396b3f122f045413de16c633570653182056fd308d316b75",1646]
+["ba74c98c0e8069d82c14dabaa3e944d6d4c031ed9afeb805a9fa5a3baa2c0604",880]
+["32fa223acf9844933b3c81e139961aaf45a81da1208e94a07dae212220f1e493",1646]
+["fc8d7a405d298ec01a0c62658801f5732adb2f3e4a6a5a521d056d8ecad95616",880]
+["5d8fbdb5aad22e15eeb128aa00f47e356e78343ad82cd20dccd79c10f7526914",1646]
+["39b9d66e6ee42d107ac2664f764c0e8ba950c0bc9c3b9f05f9247a8e467cebba",880]
+["835befb89ddff2fb03639da473ba6271bba5d0ef29d8e8c00cb7819e1f09eace",1646]
+["76e9c479e721178018293f1cfe1943a82b348496587cedd2d10277ee4ac0df89",880]
+["a4e64bfa34642fe93c6b8a0ecfe813738e8f01d96a22c30ab38988dc1d38c5b0",1646]
+["6e9a09e2d91c0419b5a2f69885dfad0eb986cde0d5623369e944f53945ad7131",880]
+["6792e4e35c4dc54bcc6390ca2e497b15572cf0dacef971871e7527bb5e98a451",1646]
+["725bb61c7d535eb482319cc55a36ceea85c0d6a896bb9a2dab9eada40e4262ae",880]
+["80200496c4344f58545403c49189e0b981328a1c41131aaf51eacc91c45b8f72",1646]
+["3d9a85d710bad3b54fcf60a00772a9c7a327e967a26e0214b4167b8a0c46ba9e",880]
+["f2104ded8aad5a7c8a6c2ec6821e35fff254dd5936806168edb00bbae4f2a562",1646]
+["6e315804a8c390d5d2594424ff1526380484dec1e11190b47dcd1b19455d8604",880]
+["e2538a3cbdef9798ca64b5fbee99486d0a6c9512813b21a54dac1b0bf5156ab5",1646]
+["0b495bbe6593325ecec8cc953a90746348a4b6afb5e6671e8f60e8f747498e7e",880]
+["042551fea8b871a312391b981ac9ceeb9acb2ddd7aa8bfff82708cbf93d4e904",1646]
+["e396bea141dee02a58b7b3d93c9b585d0d24307e87fd26ce6267fc2520d2d691",880]
+["225693bdbea8d073efd958c5ea75d3404cb2d363070e87f1c3d81088693aba42",1646]
+["c1f0faf27ba2e3d92619141a232bf3deb2e610ec26f840e9375d6a6407fed49b",880]
+["d12a26f3a9563f6de95b7df69bcd6fa85ab66710b73cf3fd61b80152cf743b60",1646]
+["c596c8f67844d7f681571b9d262496fc2ebff48d9a9dfb51093f3f2ec94a69a9",880]
+["c04e708850706518c786099a5c903bc6bf217d587821017e47eae44d234dce2a",1646]
+["50a9f803fa916dfe1f91b8844da7932def19372c5b56f190581c1fc282939d03",880]
+["60f8b2a1937cb9dedc153fb473d7df7c76719f3e27fb621bbf42fe141de61174",1646]
+["f6ee99a235e553a394623ccf01793932f610149c63f7bcf413a920c39d4ee557",880]
+["90239a77665e8b6895023a995fcc183578579699bba437628df75e12e994ef68",1646]
+["507b59955b6b09db17622d30f1e80e0c1651a44692eb391023059c11305fc0a3",880]
+["656416a2292b9fe2827fdc908ac94145c92a506c1d95587c3f930b46de19f856",1646]
+["467cfd60cd13c33bf99b538da54abc5d8a69dc2e05fe4eac1d33781d755f8c1f",880]
+["d4f3ba5273b85210a5cabf7d1f9786470bebf831a9afd717062993cf1447bcd7",1646]
+["683c556f2581363bfdfc62daf230a774ccd87bc9fa3d1a0baed52880eeb98862",880]
+["8c179d22adc68412be1d17773f9570681c7e6218a831a7fd43773f305fe3a1aa",1646]
+["733bacfd84bd7380718de6b705d25e218ee19bda6272936ad38a5878137f88cc",880]
+["e2976b08a9228862074048dddd71137a6494ba4125a60de825b3e1aab3b6eb47",1646]
+["a777d3d32dc48308cda1e82cd18f68da1ee9cab1e683ea6f3346f3a6be8ed3e1",880]
+["e4153ed10db6d1b04a8ce1a03b4511fe046c5f1bec9dad56746d05ec962149b0",1646]
+["c9302977577a76bbbcdb013afbb11942c9abfb5bc69d39564ed3cb6c91fbef6a",880]
+["d7cf798df4a5daa9a077f7df59bc7f1be3a901a4fbb431ca361da1678bec40ed",1646]
+["add8dfb399e43d5b1e34e1bb2af5d1fb901bc0770b1fd078215043bdc632257f",880]
+["b9c22e16ec88e28b46641a506508ce2304826abedcb78e919182b49fcada0f77",1646]
+["6b587aa1dde29085c69921baf8d276e7a4c168c4a5bb4fd47858422c53bc4de7",880]
+["f58da6144a6134ee97bf34bdba166cf02cba6460df1ed4180222e1d328cddb86",1646]
+["c4b5c1ac6a1336b6c3b1667f480490705214e51c1b9b821061878f68e753aa75",880]
+["85430724057f83ed0a12b18f4cde3ef9e2a85dc458de2e642864149852785b5c",1646]
+["f53443fbb4ed53d4a2c990c5b6249327013f9b362ba7384b8eb243194bc5ff04",880]
+["609bfe277dbabda57e348753d6c965ae83e4e9a77fd805a5939efd91cfd5cc96",1646]
+["883396dbb3227bbb66168f1b74b8dccb584904e68f7645d727facb26e075cbb0",880]
+["369d5f6833eab13ea2139abd675da5e61cb1f76d67c47bd1f288ea7dc70ef66a",1646]
+["9c40f2651e2602988688c6e21bf2151f748adc30713cbdefa523753c6ec9bd5a",880]
+["0283fe232a438384bc0049abc8a08dd1c6f8eef0b910a4ee87166914a5cd3450",1646]
+["24f7f5a606013915a432a2a2664a968f83208f11f1a29a668f9f2f0c0b16c94f",880]
+["97290366e21d9f2da4da15b8e0e7d8c423b4ae5bcb86c96ebabd98e246957b95",1646]
+["a22d803a5950fa9c9d8602cc12fb2cb2fdcea67a7606ba55ca556fd7cc9559cf",880]
+["7614d8389ec381bc96cd7295fa7a60f96ac9e62fb14020ece3d7a257dae12140",1646]
+["306854f92747418d56d24cca7c26bf3eb5e5896b60ae27f119744acebab311a7",880]
+["add35cfc1a6088bf3c84de603ae50d5fd30e41687c896af34b9e8019560a3f18",1646]
+["89e6e9e1dd9ba49ec8725e6ae433c0a881f9c6ee656150fdb47793591763742e",880]
+["7f9c0d68c500d6b935f1bb98b81f02032aec409609da44829f965950c4faec53",1646]
+["3891f76d23a150a8ad3a0e7d7c623c1a99b9331d69794dd362a69ce5bae47181",880]
+["6aee9d9890d01936fc3e247bccf8aef85b5b209b8e42de2bd70a31bb5cb80793",1646]
+["a769002e4af40bcfb8a7bd01de63e6f275692b361aa6eecf3ea05309647976be",880]
+["a0a81543fbd171d0c05210d4f61924d87e115b747e4392263bcaae49df37692c",1646]
+["c3912b2bb0cf26909bb55bf9dec51e425f603c0e6bcd6ced7fcfd7bf72ef3fdc",880]
+["f5670edc6aea038153fda6dc1dc9f54d20c3eb4be5e43fee0e06c12832cfd0a3",1646]
+["05fa42c168528f73a029d9ca5c5f03ee69335dd7cedd1c01950346739f485472",880]
+["a0da30994e03c8c7a7056ac4b71c0d06bc49258e24d365b959d4328f45e46a75",1646]
+["b934df1da40f40594bdc4ee0858a9b4e38961b2e0e1413463fdf3346694e127f",880]
+["1330f7286a2aa70dd38ce46470b6f497008b90743944fff31e002bb27f07e382",1646]
+["352faa51e315a8a5e4592801c0a9c0e1404286c48ecb40b8d235f0dcef39bcb5",880]
+["41866f37bdc3d328dd3dcedf11b11616419ee8534e41acfc0755db68d3837cbe",1646]
+["1bcf09509ef3a40bebc55a4d8a9e163752d5c4bc6e256fc06fb82c065893a6e7",880]
+["28cf0960bad4d4cbaabac35cbd23505ca2725f3ad489b33cafad8fd1f3a32781",1646]
+["5140f63e13e596ae910ca1c79337dcaa9054e18c7014b8fe7357cb15fe43c534",880]
+["4d1030131f1043176c919d12a277ba3b5b1839a46c9e96811b238bf4d68b2d1b",1646]
+["c64bf2244c1bf5cf133ac1f920be0fdadd31e531ad4d5c5be13d42f04659f5ee",880]
+["5c3c25e327c19f8f4fbfad6bc308ba90637ad03ea607f4e04f06a68c1c048cae",1646]
+["d49e043ac25198987c7a9cf908a56f7365c29cad9ac838450758392cb6c2f3b8",880]
+["255bedebc8b365b3c78a3b338e4e5b95b1c25c6204db3bb93d6b16f1ddd4b446",1646]
+["88f4d3bc7153a3a28807733fbeee4cf5132a9910c260c3d4dcd1295f22db27b4",880]
+["df64fca1778e1477a52b98aa59eaffaf24fe9b49996a32d55284fb00410295a6",1646]
+["85c9ad6391ed5aa37ea60dfdc0538bdd1cc75ec39d9e0777542deea4f9374403",880]
+["eb91659bbb8d62b7671822db2ffcb11bc335ebc1f4cc231e9fef95fbbe1a590a",1646]
+["305d50bcddc30f9380d9cd03e60a8067257c5f3a5ba8dec43bbe1d07d0b5b9b9",880]
+["d2b67980b68a8c93e5ac5081b823faddbf761f545289c3a0b9aab2e890914b49",1646]
+["8ffe38cc3fe69600685ddde4eae14f5cf758707eb8279e9a230f1d9713205459",880]
+["98bdbe749253c2020127e78db71ad2f9575d9847aef3591f948089990b185633",1646]
+["51278f3374fd83dc5755c7c4f32b5577d127c5d2e0996a1bad7d5d0578ba7c95",880]
+["1cf5b620f568dda555bfec3dfa244b4bb55a01a5a05e449162d6512c5bf8e4f1",1646]
+["470567ae617125661aa338141e7ab16f1ccaaa3ddf5bc5a0f88eec60571de70a",880]
+["8f83fabce1146cd637b89308935327b8ee8626c102f49447c6781872ff144cd6",1646]
+["990253aa68254b0b2dc648d934aa9c63e2b279c0b69e90493bfe6c764a6365ef",880]
+["2cb7c5880894e29b6de8bb27d6e8cdddbfbb6826acc323de3fec1accf998295a",1646]
+["2e5bd6c5cea94fb33e8acb47e356d68bf6918a48d4f418ae790e1c642fcf83a6",880]
+["c6a99b0ab43c90bb2f68a4c28bccdf2a337c542a14af7e22a306f322797a87be",1646]
+["73606fd1cc19f2f1d8c462d2c232fa473929d7513222452dfd34ca0d11b4017f",880]
+["0800d71d58b1939a4a8ac60bb29f6c63ea8ff27e8a9d51d30622398382656ee0",1646]
+["60addb76dcc0c1264d55a3b59dfd9613227341c3491ae5e58afa1d29835a535c",880]
+["ef4886973ac36dc83fe809c1f9ed97b1acd73f135239fbd3f181d554caeeab72",1646]
+["0202d40ee61c0ee7fe08bb438f3d9dbdbad2d6ca67c4533ce62b771c9470c80e",880]
+["18e1c135c83c3b0729d88d0e6d75f0fb992761f192a14ab07b38725a7524d0b3",1646]
+["7e4e511b18f457e4aa0f9ba85acdd3a52ea08ebaf0a088e9b6897b18cc8fc55e",1016]
+["f475db42784fd81a0f579cce5ec3aaf47c1f9c1034ba38a225a0f86f45a26de4",1990]
+["85642223b776a177e5aaae5c7f3b39311695ddc8c11266cf427ffcef79a4bdd3",1016]
+["052e27c23a7b14479d7ee25b69d412d05fa5b7882da90ca44b43d3c26ff3a2bc",1990]
+["4ab9447af78e9c5e1f94bf67a8bfdb4f9b713970768ffe0c97b225fe685aa158",1016]
+["53b82a421434559638724d36d398a1e2febd6312869b1c186c80b9d9ab4637f1",1990]
+["a2e70e24df170f037651724b255e41a992352e0cf6bcfba16ab4b14dec22cf9a",1016]
+["ecb596e471e6f05e73869014a4a7d4948d5815eca454c0f109dfb3cf817fc68e",1990]
+["55a6f4a17c19928fc8c4e4e39a4776c384e89d087d2d986d4352cf5e6ae2de3d",1016]
+["a2d948361f84348a76621e10e2848613694ae82373f2f9ab656335513b08002a",1990]
+["e2e1d79352e501442e9c8b27370f838b55445c8c0aabcd08d05024260432603a",1016]
+["d61637740aad6f8b7ad60c616106024b81138a342e67930c231c1e5d2ef3f0bf",1990]
+["fd368ef41e505ae1cfced0e2bc91aaec3f663b8c5964c084f121358a810aa1b8",1016]
+["2469da02fd3068354731e9185d4a5cb4bf5d79b465a6d2192eb8febf05f631fc",1990]
+["94b54f7c933fd1a9a424e5bb44dd95a18d0a03ca881bc4e8cd777ea66d04a6a2",1016]
+["e7989584e8c3eef99beab6604bd40384789f8a6b12dbbb075b41fb0ad132add2",1990]
+["289d4fbe022969fd04ad98d97cc6ea1847728eb2a8c7147b6e165401c909c955",1016]
+["a2a1ea9b81afacd0edab2b60ab2a2ba17a94a466e4be7994bd69bdd760b40df4",1990]
+["1b2df31ba0a81d24a2cbc9820e5abd2d40765ed5028a87e6569dd38d4e70d342",1016]
+["e04c16aef3fc56fe9f9124f701ae9140aa4ef92058909b0b4f35e03e9c73d803",1990]
+["db1282f2a2c43c9c4e3680e278565d6fc00900643cde365770e1746556299b99",1016]
+["44437b15d84caedbdd99760d102dc27877eb84dc24515d4584241e637673e521",1990]
+["433438c4d4582376aa996c8269e469cc63887a05683d2f3ff7290bddb452baf9",1016]
+["f7d01aa5e252aa966ea4e63e4374a4dc8dcb843572c8417d4139afe10757c19c",1990]
+["4ad818b5c63caea876d3fb9024457ff9d1cc8cd3c174265b0f8bc768870aae58",1016]
+["c28ee9035518f5c910a309a232575c452b2a702684fb03698043ba370986476c",1990]
+["543ab0ad1178624b3eb9c298ca321e2ab0a96be85be804ba7a4e3ce8a32220c2",1016]
+["47d7fe2e641922d86f4b9818de0ba6baeeb6a56b7d25836abfedfd5812fd5893",1990]
+["fd9e21c7e8f27381a13fb77930b119de5bebf428823fc83e481958541448fd78",1016]
+["e208d2e20525800bb1c6b35db81cd74b55e3ac7251d789c7b4075d893d56835d",1990]
+["c0fd34620929e99a1d5f2b96fdc93adca73a7e9c65f23dee8977574765ac8f0c",1016]
+["389ec8a8f2f82d350832f2ffd9bf834ffd0bbb9ebad32070d6055bc04db764ba",1990]
+["da276f5a0d1db068ee2079df9f9a83ec81ebfd440f43ccb9b97b8d20e657a9bb",1016]
+["918a8214dadba0121b00c599d16f9b1019153a8bbc1e6f89ffa61c058e8e6972",1990]
+["ff9e76d1610cab5b6139ae9cbd06ae235263c1e768d08c55d25d6ffad655e220",1016]
+["9eb8914e8d2929969ccd96a122156f2c12d9db26b38bd0bedd1beca8ea38aee7",1990]
+["41f1142c422b84aa318124fd12e09356b677683ff1ff9345cbe440306e9c3bcb",1016]
+["c5cfdb593837e017a9d1c9f64156a4d66ab66d3205d55af9e7b586326015ee5a",1990]
+["be28c8e276643290317a32a4da812f968dd585b415e77fefd23bfbc6f29ac894",1016]
+["c8641dfd1b8d189359cd8d704c20cc1bdcd79977ad5f4fb6720d83dfc4cc9d95",1990]
+["5478d30fc55510045ab30711d8d8d51881f1bf0d484a5933bdb820d241d22ff4",1016]
+["71cdc94a6c3bcca81c001933807b0c5315e752f931817b0bf588244e0828855e",1990]
+["59acca021ec4ff121cec2a1f54898e0843e925bd598b179f648345f2a73e2c91",1016]
+["e57cd8ef7f3de5d15cf6115b1d7dac254f60cb84626d14090fe2d8fdc78e5948",1990]
+["2513e7b2d53891f9d2b87114e1ea0e366bdf3a9458784fd6af7bafa0b63ec411",1016]
+["721abec8916b3fac71a1d8869d2323a644723cda4e54f07313fe168a6ef11916",1990]
+["a564cb0f44dac13cc222009c67f0792b8a850627a3ba40ce95301dfd932ac70f",1016]
+["16f5602e30553461b4204d378d8f1cbab0362891a905c2c8503f65271dd3f2b9",1990]
+["cc05ebb54831b7a194ce29deef8290bf115e4edf67a81c1bb57cac41d827b069",1016]
+["3d45bb5c8abac6f70d920a7ffe7daa1f232f3d4f20b7b13ee449b5c42849228d",1990]
+["4b08273baabe547013e323fae3dbb941e4e46736bbd4c88d0fdae550774ad05e",1016]
+["6bcc5e2343882f800cba779078bd3b5b0f32415a9f8c024682244c55314104a7",1990]
+["84d6b6579cd8dcf4e65cb0799d75a0ad7333d6aa40d8a46530106962c6932b15",1016]
+["06df35b5c008672383cf79236e9bbd1c7fe5d121fb0bbf45b7ee0c910e261f46",1990]
+["5b249de2c1078009f84e7953df47fc80ce5db0e0910eee700d51408163925482",1016]
+["58d6005303756bad78c1fc279ea74e7ee2e0f25a4b96a3201366df73021566c4",1990]
+["9ff89c2a3b8f50965c1b0b79e603d840a6cfb7a124b69ae40563d96a7ebe6524",1016]
+["749b94acf789e5e0a16ae99c4b1dfdcf853cabfef34348f8b60c9b94d339bbab",1990]
+["a0298b6c9077cac734a9bb8763ee643fa3d75bc9bf9134be25c0f20dd2e93cb4",1016]
+["937dcf0bc0ac3525d5f7fadeaed61beb6365970ad7ef0ec7aebddff79d056735",1990]
+["7ed87a5d11f37a227cbb05e9f87145393ba866d43d452aed391f9d7099447deb",1016]
+["6f9180d2fe40dc93f730a3e34c3fd14cbde92dfc59ad94b8f625d13d98f2dd4d",1990]
+["b6889ad5cbd1874f9c5793ea8d4c0a6338430796eb6e798b2f7302ccad1c9e18",1016]
+["c95ee54ca8127c790a0870879835eae3b2cc1648d347958e165f7882b04c4a9b",1990]
+["7d51a7f65b85808b1d8b22bd69230e9f75f298dcea0472352a5a995e131c1ada",1016]
+["fd0e8e5c1fb6fb56ef7eca91aeaa328dbced3a7883ae5b9ff38914f355e91d87",1990]
+["1bf438c710cd770b7deeb980e79ae2f651020e307c3176c1a31b074573793df4",1016]
+["f43ffaf2c50088822287eb7938cd1c29170c477efa5be1ee2cbd60d957630092",1990]
+["2b65e55fc5d7ec9feae919d3922cf6551df7f72760f2c2d39a5390187e72fd8f",1016]
+["0cc4b3440b5690b5120ca7404b31181bc1f0e4f887614df7610d7c923b8d4cb8",1990]
+["f7265a81d1f8eac3e92d4d4b0992acab49778b46e58901753223d8689e9af100",1016]
+["30c098b84d350fda57e501043cf7ca7883c92916cb6a5f2b4aa7b125cc6deabc",1990]
+["0077b9e68a3cf879095c5554c545be48449a895d407af3d6047227441641e95c",1016]
+["b9cfc230169000fcfdf2776e99691cdef39fe327d7bf1bece7dba86af43d902f",1990]
+["c279f32b770931550e15dc59e999db8ca76209e35130bc6795cddd1cefe73a3e",1016]
+["6e07f660a9883483eabe6fdc278beeaf5c329e74d2874d68e6e837327fe44ab3",1990]
+["ef94472e41f79012599e76b11c497b60a91a255ca41a72b6bcc6d93ad1dbb07b",1016]
+["a5ec7251953b2aa66b455de98cd7348fb796fd2f2a25b47a4981e47dccca21e5",1990]
+["47655bcb67b973a6270c5e983d13bc4e9f89ac3eebfd0b29aa3a444269b2b9c7",1016]
+["7b686d04ef26c29b7ef7029ca93012f98a0a817d7c7daebd0fb0fe4565c0d8d1",1990]
+["a3ea0de65e4636d61ee3a76b68e1840a96dd3df64754fb9deee49154592cde2a",1016]
+["909bf2bd8bc752123d5cb1d70248e84edacbbefb427d1dcfa59127213d50961e",1990]
+["64152ca7c96c014a4442cd8a23cb1f802c74a2765b8af51ab12678b80a3b2735",1016]
+["cc83ddf9eba8b8c31e1b9d92903b8c3487d0bb4a3be0a500905d0094920f4e0f",1990]
+["628a53879c50f8591a6b3444398905f746a831ad859cf641c8722da0e97dd2da",1016]
+["1ff551dffbc2fbfc927e6875b8ed3ec8cfa3486e52e8f1bbd7caf54b1cf3e5fa",1990]
+["b55c3d669c8b97c2acee39fb6217a8dcdd385164e4bca7f4dc68b6b8f410016f",1016]
+["5176ce7acd6b2f7f107bdae74889c522a5bac925a246839399b8c373668017c9",1990]
+["c01ddf75387ef42f60411bdc6aeb7498612e258a2f74585d937d0d6708e77c7e",1016]
+["9de6206a15a7d24f8dede7736b6fa530f039bbf3c9da2b8dbb000cefad2b1368",1990]
+["720774cd0b7cca3def6eeebee0c5cff0e1070b5aeb8fdbe754b9d451e114f7e3",1016]
+["99265650a1af6d0b8533a793ed2089cd2cf2b7106f0875804942aadd77cef1c7",1990]
+["dc63233fd4895bb8c65f9d6f50c477aea987f310ab0f08223d4b2d87db7a5037",1016]
+["72b5f4ec596a94ea3ccf94f3ba670ec5400034820af711400fb854260694eb8d",1990]
+["473eaeb721d5329dcac62382611e036a32d79f309ffeaab8b6fe6eb9ab379cd1",1016]
+["e000eafc9b0d96242296dd207a928f1e1c3c9d1719e5e060b3e3750ff408abd5",1990]
+["8894ab46df5417030dd9a393459ef528b2cae335acf0698af7f13ed088d95154",1016]
+["17576520a75eaeb97b741226c55199df72f8a38bc9b7fd43fa5074b59f8fe51a",1990]
+["95fac0b3bd8f76c8f605fae31158bdcd0a33188c88f2a4408f1148b4b789711f",1016]
+["c8768413e51964243921b8aa82cf7ae5356a5ed613a1288b7469e176a43e9b67",1990]
+["6fc67f49d47f06ee5986fb1e7fc2ab4addef56f9ebb596bbb2525498735adc59",1016]
+["92d42399fbbc11d5344591dbd831c41fc0c473f26b4ce03b5f6887794e25b3d0",1990]
+["f079ebc5089ee0dd24027d3beaac8fcc0b43a1b6e1a59c32fd0d91dcc8101e10",1016]
+["2f3164bfb015287fa98a0a0698223abdb3c29e35d08e0733c0cb489104af27fc",1990]
+["b3c1d509ba188edc2523bda49f7366be105e7b881254d487d222a37fbc18f70f",1016]
+["8198e9406f96219fd5e16a09d707399d1e0ef49e236ea97e81a942db00da0371",1990]
+["1129129c4faa5cab2b3e05eafa39cffa1c64a7d9c58210322c9c1880eb00bb82",1016]
+["be76e643afcc5b3fd55db072c57b86668ac7cbb1d47c7793a5927af01359d30c",1990]
+["774d7aceafbfacbcbdd0baa74419d5729f68d53ba856702a0cb92fa9dcbec7b6",1016]
+["649a5cee8b39a54b06c37bbe8d64e4d25339c75cd8b44893e1bca6fef0aca9b3",1990]
+["23c4cb3993f182c50d116ab554c64fe07357ac00c94a5c4419807e5042a411c8",1016]
+["b2675bdb3f2356752ea56e0ff2b5947250003de9cf63aeaf9f9b7254cf5fa59b",1990]
+["c9837d16ec197d471d1cf7552c2d2e440c1a910818d15c0ec85ac99c08271ca0",1016]
+["8be781bac8f34029b46f5eaf79b4da93d06a7ebfad325a57194b235231311141",1990]
+["1abf6fcdfbad864d4eb8bff11d244aa34d3cc07257314a2f96097e56ed3251d9",1016]
+["2f569795ec3f47f2ff50f36b487b7364d2fbbcb3f108b36e3e2ed959573799f2",1990]
+["85e84a0d5c1e64bff84035db740b8ead54fed8051f17a8581c3c295a24c9a8d7",1016]
+["2edad154d967a1dfd24a272869e82ebdeaf53cc2aa0c96e9867bb91b3abedb1b",1990]
+["be9600977144f7c218c62cd938a615ec37a1fc53e992038d90be6060751b715a",1016]
+["5e3a44464382410c4fe88c617fdce730b170099c9bbbc7e38419d3cc4f6bbb57",1990]
+["d1aed3cb0cd9e790cb54d72a4bc7b3cabc46f6ab6cd8bc9d67f49a346dcc856b",1016]
+["d0070d261d09771dec3594455f3264da2d257b5c4c38517d113d73088eb6e52a",1990]
+["1421f66803e2d7fd1d497d129ee6f03f81cb6d29fc70c1f2e638d439bc4c7bb0",1016]
+["eb83ba4ab46a041962a2efca22b7911811ba414afcf6f87bc12ac3a8b6b35597",1990]
+["ac2540ba7d203f1e4fb17d433b8b6ca676e9067a3540fb58b636f2cd6d1608a5",1016]
+["5aa650f01f9a90ea671013aba735b3858e7184f28d02e33c1f8f47399a5dc4ec",1990]
+["4a36de2f9d82f44b613db08d9c10cdc644ef1c38bb34a9254b493e779a3436ed",1016]
+["8cc257d37b2819b9947f750599ecacb36c5928212ed142d2c66fbb061512cfbc",1990]
+["9be240e34725433baf55aa12445336d55e273993c48f7196babf0bab791a628e",1016]
+["9e9fe54620e32f4d4bfd817a05a52779722707c9a2710f5b26086937cec133c3",1990]
+["83041f2f2e0e0671efa696a791b7ad68a76c057262c547713620aea0d1786ce5",1016]
+["c2a7ff40eb095e09e0fdfa079bb1ab07c98ba2ab2be33479e0aaab14f19d5b45",1990]
+["20d5eab40326b062b5c6d842c4ce9b7b5c7445154dee306c9374dd790324e879",1016]
+["aee5c708170f376e41b4b5a841aabf75191a10e19c0585a0d59706431a337962",1990]
+["793bd57e159b6033724b5f68291c7a384e9ca67e4777ccf1b3a46c8bd151ec32",1016]
+["d809596848eea1c27e9770e44a08c9069507e1db48abf6755c9626abd5737e91",1990]
+["ea096a4c09497a7914068345c856d5dee3eba08c5872d39aa79fd9d38489e87a",1016]
+["4c51b066dd9812304a9f5cef07a0e577830ebbdd6d2718aaaca0479ae3e52d01",1990]
+["a0fd281a87b61c0dc0ed8e60017ff4abdbd0608f847ab8bf38919f67596d66f4",1016]
+["02a5794310062bb599aad889b94140a2a3e0fcf2d66a3229abd8bdaa5ed575d2",1990]
+["57cb37974eb1e978dab89ceb616c51f3b605be70e104526f275de5d9ef2dfd48",1016]
+["fb90bf00a50863521cd7c6e884b8fb66a0061b1c4776fd9d1ff5b1c7f8f2e8d6",1990]
+["12358af40f23b50202b063d75f099d20e1a547a9ba415f9bf87bd13100d0486f",1016]
+["9c905871b7fc4946fc70bd329edd3407e7a5231a81262280a5762595f3eed946",1990]
+["0347268ad056178cf001f67629a006151d68d82f9277d4da78369d2f9444ce6f",1016]
+["0fbf1e90fa669d717ba15bf2d162929157e6197ba0c28ce71e1c21bd3c8014e4",1990]
+["8126fe56d22a4ab2b756abc9050f225f80e1b952ba6c65cb0421932872810190",1016]
+["8edee36637571b5d100a4e4c3d4bd5988eb296125ef6b9eef330dcf7d1c2c865",1990]
+["c6ac79ec50a8b05a7d2ccb10ba5357fd535ae6b4bf827b0d2fa39c3393db8848",1016]
+["7042c499a5df67b6cb23cdad1fc21735db7b14a2c22c688352bcab201df69208",1990]
+["50928a158d21b98914e01707e21446196c45e5cc4d539515b3f5ada78d2023a8",1016]
+["31c11b4595aa950703582ed24942f9d1867478591c02573f6983ed7588284ad6",1990]
+["fbb56439c0211024bb6a113e72306154bde19c0170682646513acb374208e05f",1016]
+["9b8db652cfc0132aafaad8080fc3cb778f7a63cea44299fa8d822e3ae85ba2bd",1990]
+["9fe82c7089cfd4ec94ff381b0b1c4bc916c23e2bbfaa69fff78032199e3976a7",1016]
+["85eec787a98893f5e6566ea3deea366aa695ba253eafe696939100372f7100fc",1990]
+["a6c1470f422ec9821eb8ab028ec4df909d67ccb222581ef9e75bab4f93d7c703",1016]
+["13af3b1d1fd5b990ef25271c89614f40b66676ad84cecbe89c7122a498cc96cd",1990]
+["adaad4d9e72462825490b43684d2ebdd457fa9923868583392fbbd1fa49e4350",1016]
+["27522829ff4799d977c89ef873e22c9d3ba2a13f4b3a96821c83e2787341a8e2",1990]
+["b7cca47f1184f005be44ca8f65869eaeace29618068c31fe8b4ace2905cb9ba4",1016]
+["f86703363e7d2f442cabc42a5ebeebd5ad479edb6c6ad775e7ea70e22574d3a3",1990]
+["e76bd4553181e73a90f1dc61ed28704ffc7db8ad9a030b7faa986277792723e4",1016]
+["a40f4e5c76715ed92f16962440a1a458fd761c9eaf4f931a84025a5cf407399b",1990]
+["d796ed024b8ed5a9142300ec3dfd42f27032cd1e114c324142eec0155167a8de",1016]
+["543106e1419eecb25ecb66e98b277c4f5bf07c6f4d2df1732efb0b2fb68b2367",1990]
+["c61c08d27808c365075506840ffbb63be1b0f28bfe80dfb19546d470aed36578",1016]
+["c1a276862c272ee9d131e0f989e57d4503b2a310edbb5052d1b2209678b4fe88",1990]
+["9c32fb84b297033c96f3a2f9010f4efd785326427c799a24fc7c5fbf3fb65ba5",1016]
+["30e2176b5f415c10d15d59db1704c282276e6418caec4e1a16543670fba6adb1",1990]
+["dae2a5bc876954c09a8aafd2df9a0cdaf56d02237db24b0aafe7ef8ba54d8f2c",1016]
+["02f2954397918170e88459530bd75b627bd6cb8a3f5276280e2ddea0531265e8",1990]
+["6cfcaca2f92abff46d1508491249b47568c8f3c6a140058e03981e5bb49551f4",1016]
+["666a2bd0b6478b527dff6d18e43fb2c2a9c1fa0a7d7b92e45b16bf0acd17979b",1990]
+["702fa09498884276cc2771e9383366eea9f7dbb929234a4c853f771341e952a4",1016]
+["bd7600612004b04de775e59eeb1145d124a47b64ebe1c8860679c760d32b1d5b",1990]
+["975fb8ab9c5690d35e7c8231df176aed925cf52483d8ac32d4300da32b6ec889",1016]
+["ae09c9a6170d97ca52c981e56abc3f0aaacc7d9922974758e8b760d87f7d2dc0",1990]
+["4a70c4acf674610ee007690fd8b50a647b3418feab6faa10ab5d82185ec0f185",1016]
+["cfa3d20100c8b7bae9e708ff11232ddda9c5814d266f95a65b58d124adf0c397",1990]
+["33b929f0859a7937a935b4f303060b6392e84d24bb4b05f773415ae274a2e2a0",1016]
+["17dd53b7a43b3d1bc738efb6fb9f8d8fae8de4fca9e92ba280ec091fe626cc5c",1990]
+["1e357999a10ec68a05c2898235486435b7b3a67adf6703ed54960c7311de0e2a",1016]
+["0a4003551de432c3cbac93ec3071820a658585e36c07c39433fcc5a04e17f7cb",1990]
+["d2e758b312819f082eee9fae80cf7bffd00992768b34c4a8fb9193ebc3b1fc2f",1016]
+["fb0f4ca1f2c8245e4fedeb0092ff46434642c1d8031651cbb960941d017a55e5",1990]
+["2e2dc68684776049afd442ae3f137df25c83c1b9e3235e4efe51d4031d08bd09",1016]
+["99d1e52af5bcf1b6174ae504813663e512234b0297194838d2a37c15f3ea73d5",1990]
+["936e5224561611685e121a67c06c609c93e23ecd620847f4e8d29843cb89e8c9",1016]
+["2e0a29996592b86c6501041506a6f991dfc7bb8ef84c2651ee2dae55eaa9cb44",1990]
+["91ad2bb1ab490676cb84bebf9c8240ab6dc5ffe0adf32c1477e3b1fb06c1e3db",1016]
+["f5636a423348f7b971c17b9e59b36751912e41a1814579a8b72949b7b66daf3d",1990]
+["3d24bd0ce43e2de5563383e4c879e309de2a7cdbe37aaaba175fe3c0a79a1506",1016]
+["472e7639ca0a24704737fad423ac07c4e35aed71a404a0ed2eebdfb29d9ba95d",1990]
+["9f0a8e26e59812850c44d5b8efa53d6475b8a1201a563276cf4da4bfb15428e3",1016]
+["5a85103bc05228af38544616a63f5e576936f962a55a75ae6cc8749238ea69c5",1990]
+["1cd749c86b5964b6143f9df5f995758e0cf9514c6e29c5058392973491b40eff",1016]
+["1fb474b154244e00f83b21d9ab6383574f244334bb352f31bbd6bedb25e33ce6",1990]
+["71d757764d08763e1bd78f638c9ef847f9ee58c7509a66d1e3aacbcf28dd5135",1016]
+["2a5610dd419118d7a6eac85f5f5fe48ff6e5a9e28ba03917291115501275a7d5",1990]
+["56b6069061f90a3414f724e88931b0b7e638a53f8f0ffabdc8a62ec98400d611",871]
+["28447130278e073cb61fd547ca94dff5b193e99683857e747226b58678fa1013",1711]
+["cdaaf2780d838e8a64995ccdc8cbed37411ee0d1bdcf77df74cc4909514400b0",871]
+["f9f4d866c7a6a2591ac1b96f8a8059daee92cea3dd4e6837dafed06bd64ee44a",1711]
+["c904804ec54c2d09a03678f290e11e8bd3aa6419e15e3606621b125014366a85",871]
+["435b19e1f8ffeb57c4ff44b730708522e53bcc26a4499d5f72cc25efb60ff7ab",1711]
+["d2cc7f6325c5c49086534f5126d41c42dce821ecc765c1375db8a7b8477e05e5",871]
+["7c84721460c2964b6b6b0239c00fef2d976470ccca69c1aa50921ed0ee0c9a42",1711]
+["a742a58611447d27af722eb319381eef5427cf8a986c6e60c643fd3c4254dc69",1007]
+["a45e30398fec5bfe4327d4f8b3bc7345c0506a5f8e595f88ddd4ef733153b7b0",2071]
+["2a0781d8b7d9f65748f45ff0c79a1fe54e805c2792d226d4846d70b58546081a",1007]
+["e5b7ac82d776cbc83cd4d589db33f212fe38aeaf94572136a1db35f4cdd2bd64",2071]
+["74e99b03380cd364d4e0d044ebd9ce49353067d9f18eafebd1cab579c48cbdd8",1007]
+["af7fc6f1af6c56ed181ae3dcdb6408fb80dd68e6e0267b2b0f1a4784c2da000b",2071]
+["7e396ab8cfbce7a3732be444cab8bd0b335722bec2504f094444577aea4b0a1c",1007]
+["9bfe8a347e410150bb04880c8f15de90f81902d74a6342e1867821fa6190220f",2071]
+["e771b92e6cf9c789c283abb04ddf2d8f6de619a5a81f7927e707dc6b12b3cfb7",1283]
+["547265622ff03dcc9e7b95e566dd30b6ee36cd838d06f6bd49b604f4b72d3be2",3126]
+["40ebfa8d2b78ee5e3e6d29a8d3e1ff6c525f7256d440ee3f7bcfe62ede89a9da",1283]
+["0a66a33368fe823f368b3bc9d47bac4a02304c8c31fdbead78d2df305e13912a",3126]
+["3d48b3710655bb94587a72e8e568c34991e8bf842a9226c0768ef68fa5354504",1283]
+["92010c92c91df8521872817418e35ad70bbbeadd7f4f9bb616fb162551da6a8c",3126]
+["c8d35835220abce46a45801ecd6615a55b5c88cf94beb0d08e4cde5048a0ee81",1283]
+["2c52308970414e02143bbe2c9938cd0fd25b8804570c63c0a73676091a3feef0",3126]
+["ffdd39427cb83181326288ccbeb51567a10dc53defff047a3e38875c2dd28be9",1283]
+["445b8ff15fea313e302405f2e2293c65c5f655c69aa5beaf4b534a1c37e42e9f",3126]
+["6a8b8f8c1a1dbe49a2aa140d600650ebd0c55931feea497c6ab55453468ac8b6",1283]
+["ca0848b7868079173cca466ac3ebdf5ecfbdb65bff90d9b0c72afd44aafc037e",3126]
+["e1ef7688a5d64343da0fff65475bacd16f62e4f627a1556d0779509acaf2b523",1283]
+["a337c0fed6402dcdc8bd3630afccc7b6a24c148240155050cacc64b6218ebefa",3126]
+["5ae73a9b81e2e7cedf37ccea7f745ddea7a6df8ba19ad7a5ebbcc60287eaeee6",1283]
+["0b9ee024b86dceba5f4e4741aea6f52b8f595888177a538a483d96e73f36e2b1",3126]
+["0f86153447ed381484962e39e39d4e7649ce9608f29e0d64a93833417080953f",1283]
+["cdc2013395a6bfe03859ae940238eabe71434ffe2881d0b89d83813a8c5ec811",3126]
+["a48c5671e093b52270e7d093b349c87183179184ebb1702fb75596cf917a3da9",1283]
+["98bad0261d29f6f29c04a70fe1b4e370af0c271638e7df50c85c0f507dba1d23",3126]
+["f405e612a6caa56566696740d42a20dbb10d2a4cb4f5e4c650a10a8668c10d0d",1283]
+["c29eb06fb728f0ad61af138028d8899c7a126eee639b57705f18a923496ede47",3126]
+["a2311e0472e545e81aec8e453b72ebb4356e42c71a37b95a24085f2efe1703e4",1283]
+["2166495841146f0a7f0a66b85a1e62cfea8f5b58da64bfa33f9bc51060bd65f4",3126]
+["f9a1c12bea14f7dbff5520487a0a9c27d9927a939d217a2a518f465f0740bedd",1283]
+["0606ffbe41e835758bc96fc9e2e06e6defb4acfc2777cbc2b9e71d6bdf82e94d",3126]
+["f2916bdb18fb15147daa4ccb27d3c265d059351a03cfc51499d853077c8240cf",1283]
+["c4fb37cf3aa0968da734bae9aa1563c7abb002a58727bbfb9aad875334081ff6",3126]
+["8ae85bf85f3613e0b7c7428ebefd94140b8ed24cc24ffda766cbdc77a846ffec",1283]
+["2d4a887c10cf7f21f3272c75d541a4950b62f41bd1ecd72bc5306f0b9f5604ea",3126]
+["831c1b3e40a4bd756175f8cd30a2a6a0ae3d121f5eeb3d4a7d2856dd850a07ce",1283]
+["7039b145fd890b923390881cf5ddcd61833eb79cedaecd155eef9586ef7bafe8",3126]
+["0523c0b5dd79c82a65b9e6de30d483a09adc87e247f5fd0bd8d180f289680cc4",1283]
+["8cfd70d364ed9a0e0c1ef33b3ee40356982820b39d0861834cd3a80c1a82259e",3126]
+["5bfb0e4a7295667d9bfe0d2c31f531ec7bc7648d547eaa43007174a2242230e4",1283]
+["2e825e87e60c2586e6e5def96fc863a3b7bddc17f7fcb3aa3b409aaeec36109b",3126]
+["7b4b24519742f4f8a99e880c53e0206a8f7cd68c66670717d61e28b37c4bd59c",1283]
+["36e6ac6a798ff5bb8864db3b0bb8e2bfae012fcd67bd2693f8085f5c4b718a83",3126]
+["17e93b3221829828523f7a53073d1869bc4270023cd672e6eaa4c3ffd03e77b9",1283]
+["8f34a7f6f8630827739c490b57f0ce74a7dab78746e2c93cb4a9335d1b3d5f85",3126]
+["67a66385bf390039769ed3ecc974850d876153b1b921293b110659f464844f83",1283]
+["85322d93411ed73a5a179ff8c8be07e6d91e7873640e359f69efea62e23ff763",3126]
+["2552e8b03087aab5502492e2d6efdd7a05883751d7cf7fa2d3a4ab85d16324ce",1283]
+["4e2af0691ce7f711ec2f71ec6db3f71eef4b8b3d90339d7dd43455457450adfc",3126]
+["76b9754d841351cff1b757cbad7e3957f69b0412dc8061572f5e6321abcaaf74",1283]
+["a00375f2fc6d4209d9a9dae129dae50cbc863f0265d31f0c3b8560f2c617196f",3126]
+["4cebcdb9313d47a8fd1df6d33a83bb69691df1657e728af3cd6013d879d63e4e",1283]
+["38fde534bac984c56ed2dbd1b6763d9ff41d6ac675f711357635e218b6cf6831",3126]
+["318cd73786f74430632766808a635ae500accb78943b6a53a963fb9192263d6c",1283]
+["8cb401c8e4711e9b23a492af945cee1cea0e83b60d56dd16516819293e0c6c94",3126]
+["579779ce6802b2581b9ccb7f9cc1a15a78869eec4370ffeec5bfcdf0b540e2fc",1283]
+["2086551df9ad5b3c508d79ee7eeb9f5175de484f4dfc229e2b2e754b482e17dd",3126]
+["2f741079088d1019763780defb90a4f714feb4854da4b014672a679e7e0e5157",1283]
+["d92a21d0f83ffbd89ed326781539a5103ef514d0e737dec8624a3e551a5792e1",3126]
+["e9580f09eb79b73055065fbe46bc59163818f90eb5bb6ec67ca4494d5ec14de0",1283]
+["0dd9245c5beb8b458797e839d9d078b8cb96924d8a68ce4b1fbd3c3afb887d7e",3126]
+["902ba5c5688961f9077a1f1ed1f3458fd61c2ac9af71472591da873a0a7091ed",1283]
+["ccce0b488a18bc28a2318599dd8ab2c71d0550cd1c7d138dc19d93659b4a4244",3126]
+["308eefb2b70945a38a932cd42be3a4710c877d195113ca1a0778a90c5a683bb2",1283]
+["8d0cdd9beedb17743fb8476d95bf1b8449fd7143e4c4a5981530561738edd837",3126]
+["08817f25cabef0a5cd0ad4844c71dc5379bd7e24b38b4df4fdb510cd0073e98d",1283]
+["486b886e80752ca7d5c43ac7275c4bddaaf99767d7dc5c63d915ad836ffd736a",3126]
+["b9b48ced71238b0f041eaf7575e24abd300f6463f60aeaad703a1453238657f0",1283]
+["3a99081a3c2a9256405aaa0e8e34fa21e32f7790be5368bba239ad0f2e6685bf",3126]
+["c0c343b55afba641a32f78ec1d5d19397e08c161852d68fdfc6bea380b22acbb",1283]
+["96df2d2509e07fc16d0b0b8cc7d65eeea89cfa2f7988236ce3b219af3fa35ef3",3126]
+["76146163351055ea333f8e21ca7707226f675cd155db219e8144d9e4c4b4143d",1283]
+["af02ae0bcee089da078bb1d13014f667b3f740d1123934f7b6611dd174cf1859",3126]
+["a5c5f1ecee0a9d9713eb02e52497304ff92652eee234d0653a174ffe9cdf6f83",1283]
+["a759d1f2937860841e0f08875c9c534a1ec7cec4c55cf1416ea0b619d95244c8",3126]
+["c73bf581b33af42aec2d97a75c6b48bd8e182adeb85a455b2c6aa782fff415ba",1283]
+["bcd2d569b6835f9d3e3ba3adbd497c40ae1aae85893fd7327dd6fd67e0306636",3126]
+["4653ae93c091bdf55feefa59a577f7d07c623920c5b541c1749ac8d5c9c54ed2",1283]
+["df44e15a24409baba32600652fb26e55013ddf90f20ba8768ee4592666abaa01",3126]
+["34a9af29e56b4087685b68826a60b82539b4d63d09fdcde5d7d29fd5d31bfb3a",1283]
+["e1925c03c7e7381ba00835b65ee8c27fe0acaf22a8d7a461aeb45463f2ed8a2c",3126]
+["6b7beae7dd40bd6815c6d8b1efab80c36e7b2b5be13b9c8e9476ca659740314a",1283]
+["b3d2c2c26b9ed857e324756731592d9884ead465c3b803f6ff6ed577154f9cfa",3126]
+["fcabe5a7daf412449dc96fe70743fc0fb7f9ee3956814252de3767e4d0c368d5",1283]
+["13e2bd62ae18c9f2b2701f6bc5d6b99bccedf0b934ac1a8056dd8f80b30c3d5c",3126]
+["87837f1d996ce2c7bcc34ff80fcfddcaf309428fa53fb980cf35fde324ecd849",1283]
+["edb00bf0d673b9b37865088e3e8264e694a3f918f1d5af8fa4c64aac71f1db2e",3126]
+["8a7854c24a5f196b01609f23dca9cab15f6b0bd248ff655967ea6d5ece43c82c",1283]
+["8600d0f6c12a63c58a7a1cdf914a537c06e4519fbf91013c8132433bf3691257",3126]
+["ec7e141cc2a0c70dd6dd383fa71aaa8a14c3c108cccebc9aeced8ec631c1ec82",1283]
+["bcea464190dda3a0b76d1079bb271a08282c4e633b2d8909016175387be04792",3126]
+["d9341c82e3726a8265206a899931b5c9887399dff79e50613688ef32f3bf06c3",1283]
+["64e0941336132243a09dee6329e06c1c7e6293cb804263de194054726fd45178",3126]
+["9bcaf568b1d1d190f77256435a3c80bbf64bfe9e22aa7703f1caa7f43eef75fc",1283]
+["b9dc1c266fd081ae5be25f7e8616c154fe5a6879687abb14473faf4f688a075c",3126]
+["4fc2b3562936670ce6d8e897ffd574724108ec7bd7284d62338c58314b876de1",1283]
+["5867ceaa60830ac850d72f7458143d258f2df3bf9f8012cde15c74ba9fe0e7da",3126]
+["fd3f022c92c2633c79e20031cd22d0a691d8c8117c292175a5a057df472bcf04",1283]
+["a0faa9136e66f2d3293b96aac60c1f195a4a990221e290a468532efc15b800fe",3126]
+["3af4e255485925134f896abebe940ef77862978d5a8551ad63dbec6b39de9edd",1283]
+["c296b68512e6a656f0035378fa3a24ab1083ce3a857b54f760c3546ae5504474",3126]
+["9135523d06fa185a434b6cf5cb9d5a7c20b35b82ae1b98523246562b91fced73",1283]
+["9838281b69080a5af0e29731cd35416fc58c4ee92c13433c78d20c81ca776341",3126]
+["9fe912b73a5383923d0cdf0f9a65b8c1448b27ddee3b9472f612566a0eea975d",1283]
+["71ded896c5e6694f9e93fd86c7a7b43e3de7595988a436226aeb0a37de70aa40",3126]
+["097872ec2a31bd300b7a48e8079b29b3a1d716407d99544a778b4dffe2174087",1283]
+["35a441c94d399c060c307bbd213d71fa687be5af170c80e8a7695dfe6ef63d1c",3126]
+["6a8159cf7cb116451c036217f3599d882c7e8428d977cda199e35a4f208041f8",1283]
+["4035f97bc6fc23c30e0be7dcce2df9e239e04f50d1e3686943ca067e4b7d203a",3126]
+["15fe30cf617676c969ab887a1a82d0d210e9715f46289780e750d730324a600d",1283]
+["6c716e25ca46234a0145533bb96f5662050637182c136d9a10e0ca68fea751ad",3126]
+["fd45c95d9c46e22473ed95876de618ba923e8e9697605f5874b93d272c5790c3",1283]
+["ad0934cf6e678dbc90fbeeef64503b40b5fd692e0db969c574c9dc5aa8035230",3126]
+["0a45f324626405d32db9093433e7d0ac52eb21e4c05a64d98928495339a69844",1283]
+["768e5f9f2ff7206a50b9b6ae3265655c8572a3d0ce4abbe882e4aad87640f625",3126]
+["346b2d4e4769ee894310bf5d2b3b18b155004338d9845e044159b96e204a242d",1283]
+["78d16a2ab6d30ef77391baf8e84da3104312cbdec93f32a3a77b65490f4d454c",3126]
+["3a046caf1c82c9b521534d7a37d9a087ef6f116bea794ad1456eb6b41af9c936",1283]
+["10dbcc2168b8710ca5c0badf41fdd49a53c38bc2215dd499e9101eb94b0baefc",3126]
+["f3f1ee870fd36acd74930a50a98e3019cf24b2bfa7346efe2cbf51c631110162",1283]
+["252e3972003cc04cbd8b770f0c726d8520aaccc5b9d3e4e8ac69f3a7346f12ae",3126]
+["7d9adbaeb34c53be58a60ecde45703b4c982be84a7e93b4bf6791c93b09ca6f8",1283]
+["abf26049cd96f997d62f41942e6280fc83dfaf55a19683439ac2fe3fb3b8e90f",3126]
+["c15f41e64fbaa25efdbf3856454f6f16f9a0823ceac66db2aba1386915dd243c",1283]
+["5afaf00154826ff7beda205548483edde8576b2a6969e4858b79a880df2411ba",3126]
+["37c1f824c045aaa5346471d58cf955d2bbe607bf99597c8c4bea31103b1931a7",1283]
+["1a602a92ab80d2bf25bb7fe9b3f73e3dfc05abd83efbb16a715b76cdfe38b37f",3126]
+["6c5a3d82ac7330847f34e184f53b0f04f02a3529dc5fbe0d3ca24d1d7085bef3",1283]
+["0029085017f99f873a878afc4749dbe909e81c6f10f25f7798f1e31602e5303e",3126]
+["59397930c5b4449717511cf5eabd04f04e1dec7b0c342df0d34c015e3caf43f8",1283]
+["246331d74945ae094f5ba94e2b09d87bf999082ad9684789ddc6d67e0702da3f",3126]
+["e4c92f02c1096729f11049f4f38d86b98af0ac6b52aac8414b227d590794dc8f",1283]
+["e31ee36769b7dc5693f9315c30aad4a7cb68a6adff501b3b8203710d4abf6b82",3126]
+["a91628ed4ea5308f28d346b0147d087cdfd5e64076fb534cf5ff26df3cc04e86",1283]
+["2df12684c50ce8b42b23ff00669dd5d2f551749ea94a0d2d9f0ca09869112afa",3126]
+["b67a7a486853a36ecd026d599c88ebecbc6527550676bfba5e8c31ef0a781d09",1283]
+["0119be8b35cdd050535c8c0b1634b7ac2d3c1383774a36eed49b67f7932dab20",3126]
+["01beed40f9c624e9fc293b8462c137232f98ee9de3ffb9db6d270e516e48780a",1283]
+["9c6444aef96960232cd7c46fe0430e98d4a4b7a380aac35ebf9c3b7ae5b31447",3126]
+["5cbe401c084a7f636a365cf33fc3d50782bf520547e27a2e2f2fbb68cb347ea8",1283]
+["92ba5b16b30939bca9185abf19a7ce4f4433a2109aee05f0e919c7c31d80d989",3126]
+["1bd10f1902e6c97cfbf1aa095dfed01dd7a4362eaa12c652eb986d3c34a113cc",1283]
+["d4b415c0e0c7b6a0e884e4f921076e7d3910ab73341f874c7a3e525c3bb4f74b",3126]
+["2498ff10da6aa7175596281a14ab7b3a83957fe12f4f05f50edac1377d13b1f9",1283]
+["337c52ee440298a0aafff562824c3020d315681f2d149e6afb8300b3302b92fd",3126]
+["c5dde42ae5d90e00aa5d9b624d04930ad566c2177d461a1e4384c9a23da6218e",1283]
+["2ed59dc9e800789aba30aa09141a74a37a1960a3c081adc9c676dc0f65a3801a",3126]
+["0f3f35098066de25a9a4601b96f53801c0c1c3b0157510d7a21ebc18196e1351",1283]
+["1cd917daf6484a29568a2683d923e0f275ca03db6b4f8f3d8e3c298d247e12e2",3126]
+["60a6ef3c616bdfddf3ea362af468ba24ef360a343e562f9464bad9fb51b9dc4a",1283]
+["9b5b64b0671558a7dc0e73406e7ac3538dc21f7c92e183779ca20e883bc2161a",3126]
+["47c68590a236d7584ee0b06e4c8daa011c0c8d27da8a35726adcc28c1bd49025",1283]
+["6c54cff08d3b5c7ba30abb7ec541514a76e8043b0e2eb2c587df2c965e41dece",3126]
+["1b54b552e5bac560271f5e61816822179d7922d86ffdb7f385828d853ab4b2a9",1283]
+["4fcfbf684d88994f177677c250165c2b63d8e42f1616d6573a926d8e705badd2",3126]
+["f5bb801d6609d7f85d84d31a48c57e1d9166ff5a08fd08eaae1c67a1e8564b35",1283]
+["0b464a6256c643e52af224172c9240779c2213677a59344050369cf73cf69ba2",3126]
+["e28f906b6f0cd605fbfea9349c3c83d6a131f3a7bfa51eb87aafe18ffd2c8ef8",1283]
+["eccde998bbf0cd14307853abbbd7f8e24c794452be31ce69d4ca7b60362d3e84",3126]
+["e95bcace484f10ac6e323231a1ec00f1f725460afa4ba877548c6bccf252125b",1283]
+["329b1b5adefe14cb28f61b20fe78ab93c58526d6b28569227af77f91e08cc4f6",3126]
+["a10d6a28ebf89b8707f2209724d11ff5af235c55a49411bebf1a4dbf61a0f486",1283]
+["1ae80f6476484cf9602e362c5835166618d892e1e68714ffd55c1232f61baef7",3126]
+["27ba52a8dbed0c8faed4a2f2edcf0e1ca9bc1772f2662a79be737682b0da74bd",1283]
+["77ce1bc8fcd01058f6175b9531131cbda8706e54b67f5f3eefc98e7555a38350",3126]
+["5b8c9230fe42f9039a498986a521f38642e9f6169865ce1745792fecd46bfcbd",1283]
+["e768a6c511c5bd16103f467c2e34e93b86cf38151e8be56fbbc4778e68baad06",3126]
+["5d765e990c9986015dbe11445fa9a40377b868270b6ae5b60a4a4f7d04574921",1283]
+["007f8fe203c02fa7acbfa44590aa9f4f0439b3dc9593af0a82acbd9f04a73a35",3126]
+["6a45eb3464ecd1a85a2234cc8e40acd06ec3d2d0c4ba103f5ca2ce19c13d2bd6",1283]
+["588d848cd40cf4867c876a51ce2bd9d22650e9abe91f087a2b7f763a10c56d14",3126]
+["ba9bfb8adf8167b30378cfd2c42a21dcbbb436ce4dbbefe07af97befe130d4bc",1283]
+["ad654b3dc86ed10adc1bde6d8e6f0445645e6144b4154c1d0a326a637004963b",3126]
+["6bf914e6129cee3101771ed4a05de1e2cd4953c27b13e5c4fb992f04a75b7425",1283]
+["29bf69a13c686bb9c5cf6363d4fd1f2546a097e93290aac246ac546a709acf90",3126]
+["0a51d6ed894888db780c64cd2374ff8d80ccf52e6d8fc7562015d5cd1c29257b",1283]
+["7c0ec952132c40054bf9b0549e272810d5d7cd984835f4f17c575d502d1d1e34",3126]
+["5babc518a46f54113b53caa49bf7372a44099299ac5e8594de15fa70bdd135a6",1283]
+["995d7851504c2b76c91fd69979374e818b3b45ce085fa7a20911e4fd6354461e",3126]
+["4292a65cdad00ecce14e9d5b0e0b175ed542f250fd481c46b7495fb7688fa5cc",1283]
+["2ebaf0e62dcb2f851e786845796d47353cc41703aa113d90d09f83f8c9bca4df",3126]
+["6720986e660a155db2bad3c9a4662f0ac74c591305b8e4223f399dbfc45f8f37",1283]
+["698d789df33f7c08a5dd1c1bc1ee9a7d91086a7492132cfb8d8fe9ff7a88496a",3126]
+["e4ab053129e51e1a0cf7007eb88b9f9b142eebb8134dffabbfabd8312bc1cbf9",1283]
+["2a9de79fe88088983ee8244f79060d6c8213de2a697fdefedfe2ad2e3a42c3f4",3126]
+["bdf24ceae5811a889fab613314551696fc67fb5759902b617c378d6ecf9407fe",1283]
+["bd4daad42b22eeb835cad453eb17575f26af9d19b2b95044ba3e1e201cc3c0db",3126]
+["4ed0527c05ff510c234e03b5803ec14fe3350252614eedc4eaf97fafcc36f12c",1283]
+["7b9ede30c3a7ebb1b8bf6776f01fa07c9b8899806a7371f65ac242128cab0c60",3126]
+["74e69e1a45f838a290014c43f5637b33a3eb668c82ef520e43d1eccf566507fe",1283]
+["b225add63105c978e0f8a9b67fa1e16cc273a8230a2f2d86bee1e7bfdc8a673c",3126]
+["93234c0dcec1efbd4a2e004442cb957c3a6646b64d991484c9a005f407ade219",1283]
+["7b03f5dd97360d6be8f2fb7316cf164486b828d7005a2501838b450645831fc8",3126]
+["421a86d3821dc4bdbb8216996472de12966a19a6a882e993c3a799a335c4700e",1283]
+["637a7f2b8cc49cbc4cf5d7205ed4df73b30ce7f24ad1075352cfeb0af0ff0c7f",3126]
+["de91c7d45e732fb67edeac7f62702f46af0150be294f6b9180d89b250cde8e2f",1283]
+["1c91772d82175448e9caf109a8c551e3de3d494061a9e8a1e1b69dfe73d4bda9",3126]
+["220f2c952b696bd22ba5f99126aaf6b118e43a9e27a8796323e347a73e8ffe22",1283]
+["f92f56f5a3b136cb8dffdb244863beb13e990aa59919369998b5aafb9d59877f",3126]
+["2f9e7a86f51978274e724073c569f5ff305529ddc8b0319df0cfc3f20459ac3b",1283]
+["0989a9db813da5ad80300344b9f5f02316a85606d15475d6029af63456adc92a",3126]
+["976c9ecd5dc868834442e8b978857cf9209854c7b6c8bf795f40597479a0eb2c",1283]
+["06a0bf2e4f075c60c46928840d2bdf6b6f80ff8a51898f69c650bc53c936a0f4",3126]
+["c4a306581e91e470fbf33053942daebd13b0f7f9a899c881672d12b5df93fa83",1283]
+["5659d9b8a3f1c3e551bc6893a2e905f9bdc07068f5094097e6d0999ce7b7ee01",3126]
+["ae288b39d2743ddddb0061ea87c5d58eda8095c4502a2beba0749b0eb59909d2",877]
+["adf7d254dfae7b9f98dcb3cfd4ecf683e32fc92395275ef29f5ba9449b31b13d",1467]
+["7d62d340a88c531db730249c3abaa6423476ca6a4c97f6d0be8520f92ea2a8ff",897]
+["7c4202f6c18358895e46113d46d4a79b788d38c8a4c9ffb7a23b62d748d0c4c4",1657]
+["e9335091e5a8ffb7209147f27405db0398ad6b03c9c1963c21770484361f5cde",880]
+["6f0d702c8cc611310128a3940330e64e4f00a4c5312002f71fcf5d9b997d9c4f",1521]
+["4de22f05e7eea52dae46563a87cb6279ba950b77945e6628671ee39bb3944cd0",900]
+["1e1d0611b736c4a0d0cccbe84f09d9e54b30214b0335da0edaf4020dc5334dba",1711]
+["fc0b615ddc189e9ea55d3068ea7a76bcb65bcc2a3e385a62565a6b8045cd0805",952]
+["e272e0a7902bf34f761fb71aa41498739474dbf2ee6d0f663014d1759ff116bc",1803]
+["f499969c2d332447dde850f675581697cb8f391a17055f6579fe9ace8c94db4b",972]
+["b81741e58d933072f76758345b65674c4909984874458b1c8eafa531e54e70e1",2053]
+["5a52d7223b05f19f1980f80dd2dc2243085be77debe2481237bf00c7d61b938c",955]
+["54294311311f1e8d77eaf863464ec20e938c09ed884629b8ad59f201c320bd5d",1875]
+["15f5f83986213cc9e5bcc432e8959aa109fba2bf37c46923ac7d7dcac621b83f",975]
+["f234324ae1b9a0b3053c59321ce602d845283fddcfbd1517b76a0fe44820ff27",2125]
+["d26a0487965c072cf164b5be21e08d3e90495d7ba0ca874d7a8a0ec31e0ea24a",877]
+["86324e9701b023d5a84d060455ca1c29330ff8ab9f88d165e631f43ba06fcc89",1467]
+["afb0303ff3dfcbf5b909968c0fd255a3753eecf8bc63d437cf34d16f46e0a1e2",880]
+["eab33cb859ce7dc8393f357bd79d1ace8258bc58569734f63b31bc1333079801",1521]
+["0c973b93ba8e404b4088154dd3a153d8385ca1965e70d2ebb8a22df1ea2965b1",900]
+["0848a7f4db7cd6615473622e5ff60725a658cc32a62b8c334c9e19151e15eace",1711]
+["076bf1fca964e82bfd5127c50383b044199812c56ae8e34f70736f093cf1b663",897]
+["60ddcffe7f6db67629dc137079787911ae15991145bbfb487c8b3af2e0bda71c",1657]
+["74c3bb52c90c0e2e7f8621f1b179e5bd56b0889c9ddd29a808fe9d767b1823a5",952]
+["863f1984fb8082c00f71181be68664fa73a8babacfd97b89ab22fa389cf8ab3d",1803]
+["20412ec670879706e18290ec6b9913a1ac82d675205eeb3b5f8cc04fbf2d169c",955]
+["3e751138b016ce9161f9938feb2a0d94e005a459e9defe4d6af583772ad096e8",1875]
+["4b3092cdd31cda7df9aaec2a7fbd1517997cf0bf44ff6798be6bd602df6d8fa8",975]
+["8e040b508dab6884e5cdd8cf1255504951f9a66baaa9974f3f92f4d6dd4211de",2125]
+["fae6a491e70df35437a3e3da00e5e4b98bf3b48eec0ff0e5b866f83302037635",972]
+["eb07f8a8fe09cd4aedb5631d7fb0c0588d3db048f82af8733fb623acf0542753",2053]
+["ed6381455a83baab63a64c7d722f44fe3a5548b6a24b99c962a24fff640ea9a3",921]
+["733c50e65f4c28011cc0fb54d6e1262e8a7ac2be5685295b49097c9110e430a0",1488]
+["c7f73dd305b83cafd5d3a57a11493fdb0051f554b2ef947d36757a3558fbdb3a",952]
+["d29ddbea8e87b3f9ab97bb75263b0b6abc723133b80fdbb550d3a2a6918f39b8",1803]
+["e3d617550d7eb12e915f599e13ed3ca70f9c0f58b260331ac0a47dd4834aba80",972]
+["4de26dfa267e44ec6c54b32c3a8b735ffbb879a2b89f283a320e62bdd9ffb2c8",2053]
+["e3ff2099f1c0dff833a1b053fcc808898637fca4fff1a272fed41f67b667407b",955]
+["e6523d69addc0cccb0cb034e65ed7898fcc26d1b0540d51eca8e6d24f357aac8",1875]
+["79e7c4735e11f0432e6fa9059b205ae7a9f3572c7e174328fd5093dd049c7b44",975]
+["6da025ba1e5b585d7c03325426717c34a572ec3c04fddcccce60cc44d95d8354",2125]
+["ead0fcc6f6518bb688c10e08246aa6c310484db99c38e10fa4fb074de7827d42",952]
+["5c9bbfd470d227ee397d5c640c2489189b41c3d1df5ad7c5a6cf05c6fc53016d",1803]
+["bb09e1ce03738e304274749dba1f5b1f63bbc7b1dc301e95ac5d3b256722120c",955]
+["9cbe81b3179d4208600e4ddca19cf74c400fea82acc1c19235f5df7a1fca47bc",1875]
+["348292f1fd03dbefa0375e33fcbdc549ed5beae29eccb9199b72eca99ca38fbd",975]
+["b41846a03c940aef9fbc5fd9b941be10045852876bc75f2a57a96c744ed955f2",2125]
+["4d631fea825433ddb3921b185bb8fd8263f530cea33ea85eb09d0cb243590339",972]
+["fa53210ff083cdc3543b9ffdabe04aa3f881430d2ed181bedf9e87dc5c2f4142",2053]
+["5791761b97971fc9ae42153f3e88487bb463cdcb11dc531561c9e56cef3b9b75",785]
+["e893dbbfbfa7a38167d50e29c02a6b5cba468c3f123613ad2a6be51be9e2455b",1155]
+["449a4c53a875f882794c093ee66b42bb83c74716c65ed06507cbeca7d0de8be5",921]
+["50859192665fa0471617547afa4e8a7faa49d45b8c7c525c7962e9cd600a3ffb",1488]
+["a0b0a5ee91378a7ecba7c6434823fae01c237b1d22f9f10c7f1da21c413cc92c",918]
+["959dcdec3481bc1f41dbcade621e240ace8742eebcdab7c4bd1b7d3c8e9663e6",1468]
+["468a2b7a04b19fbc51db32f539a5f9c63d40d9d3829f91d32ce1d1380d5905ce",996]
+["54991d74b5cb221c5ada0cb9ca72275bbd9297b73025ec02f51e69d37c732f07",1830]
+["7088d990c491bf7f1ef82c787933b78b8a992b58a1358aa0d9aaec50078c413b",778]
+["0e8f75dec970be085caa402c2c31272f977467aca22cd11414d28548d40be1ed",1161]
+["1f0d67a91b2e065ad5476b17e9a3871ce5407b7cd883b04a3c734192cfcf1696",778]
+["adbad60fcf83370cfdc421d1c3b7433a54e58c6a1ae4a2bf3528047f9d028c26",1161]
+["e64ae450191e0fb5d6f5d9867ab398016642a59e5fbe88b120e58755311e1ac7",758]
+["cbb592e75860dde04467220cb44781ebf0c358e25661cb3916cba8def99a08a9",1031]
+["08137a2e1f65048a717c1b943e594cb65af3e4dd232d04c66c2a59e5473a6477",778]
+["1263c69d977925ac8e8f6d302c741ddc1363066ab7c3d8b1f787dea696af5c76",1161]
+["abde109773a46fcc3d2fc69c26bcc2697e6a6f1fa6793006badfb3168533be2f",761]
+["964f23f38200d48da655927aaa32bfdfa08c750aa2222d745fe8896a68a13945",1067]
+["38ffad94bd8cb0cc623ca3337e8dc5687b766a4427767bd890f190fa4b2f0f70",781]
+["c3f41913257a8af55b48be035515747d496284f3786725fd2f0005ef5343144c",1197]
+["1fe4ba43b8085af58d1ec2f1ba1b5b7edc3e4e450b95e779027cb84ae5c36e80",758]
+["85c5703eb6c6b375659665b783c2c95b96bff137a4028183d31b6975fc2c4169",1031]
+["ac543c39b62b51bf5ab691c79f3b795bf27d1259b4d6f0dce8b2b90569206db6",761]
+["528b4fcafcadbfbd8687fa372c61f3cde73f01010a0c9625edd2e6a05e8fceff",1067]
+["ca39663a4d2e79dc9c73da63633711ece18f04e14c1181a9918079998506f5be",781]
+["833de77634e4d37a131366169939f7109f74139fbf76f0c838f87a614f20a2d9",1197]
+["914caa493060ccf84406ee6bb9a083a2bd65953a6243252e6481901e6dbdcce5",778]
+["58d2ae526bfce398db1ab50b4d8599d4566462d6a3f24a9475b6162a537b2758",1161]
+["664100750d6123c0e603ffedfa92bbf4e92a0277d58c79890cfec50848039be2",758]
+["a6725960467482450ec77205d10e250fe0b7e6fa74d498ec3d176e7af3acb71f",1031]
+["00a16ad612e10a4cc01d62f863aec86638d9b8b806d30fa3f06cd1f1fa57e1d9",778]
+["39fe379f494574c602426c2c712bfd0e24255621f31f6ab67f3d1dd122a96e36",1161]
+["a09fa02f8138848b53c97503514a62a188befbc1d47dd4398677cc4ef3cbd4bc",761]
+["b896e364c0a2f5e8d0ef79aefc3ea9bdee09ae7f021f360cd5f0b8d1aeab0bef",1067]
+["8ea59c49480ca5494b8348e3a6b7deaf437863f85022c16f8a8b9cbc4d8952a3",781]
+["9c3a05e12b28c48fa04d41d2715d61a2f18bff35ad81c593740d9a0db931affb",1197]
+["24ce855001eba0d05f30c3619fd9151a83e4831f76f304469d8b4378f24789fe",758]
+["e762e175e703e645c978ae4af8da11cf19235b7bd9b727052ef8a6793f769c72",1031]
+["0a680709b35b13470bc6db649606eb3fb47267394d982b7eb604c64bdfa25b06",761]
+["d690f657b5b18f1d1761e6f44cf4862d8da03a03cd8a61fd3211f93f2ce915da",1067]
+["3827102d25e63d404baf4aa191d4db238f11c70d94ee567a630391fcd977fb20",781]
+["2c7e6c2e1cc21864c57b5c7ed588f57af0ecde46594aa4b1ec4862a5d113ac84",1197]
+["d8500cd27fbc40521595e61ec86d703edfef7e5f5b33e7a8e3686f883956466a",778]
+["ae589c4808e6d306be34f2e7ee8d0c3791a810fc81a91f99523ae79d99615310",1161]
+["2fcb0bbdfe0b938132293944e5d9bd075ec2d06dc9c5b43e393fb28521d9cebf",797]
+["116023da0a7ccdcfc4e7301bf81996a2d1ea53ccf944688f06fcac70769c931d",1001]
+["b6eb1ff2d897f35db262b6b1ff9ef1e8da9382bb5f71db235f74ebaf6ca0aeb8",1072]
+["f6a80a1c837fb6788d4cd5fb158a8653d70c444d23e13df00c6480ebebfca400",2602]
+["030837301698cb71ac88c553af451fb7b3d4dc7b1fc3d537bc671d5f215edc02",1092]
+["fd398eb05a830bab46668de56333d6fa6c7b689454666c63401595d6285a2eb1",2912]
+["e907c8e44e165073e67b424c5b16dd74756172fa1d8b2e1556aeae173e94407a",1075]
+["08889035d9473dd83f162990e03092e337331c9dfda50fa728985aa56ef09236",2692]
+["f5bb6ca3926b243a010ec0475660a0c9f1fbe3f0c44ba71a38f4032cde33c94c",1095]
+["99bf8f0160216bce060d31b42623bdff25deefa1eddd57370e3d629836854113",3002]
+["9f06d05393d6407dcb2631fba13e2f46fc9d337285a0f041cf86d9a703db05e9",1072]
+["0d255903511dbde0e79c931842da0617cdd7a689934e652044085a2e348e4744",2602]
+["a28956978e4b2bc5234feef7a3fec58275cbe15f8c9cea1db9b56c91f04801a0",1075]
+["d059ef564846ee281705386d432ef024c2d0f45ba704bc0ba8c999f3c8540e22",2692]
+["ab2c0ad033f88dd955a21632ff163447e3ea343121ce7a7850eb9522aa0d5d61",1095]
+["dde06eab816cee05fc1e889f50dd4ba1ea966a9c7866d5baf5d4f4ddba5b4412",3002]
+["50d7ddc0c26612ba24e7a9b0f032271fe788a2ede6885ac743231abce3864c72",1092]
+["eb5fe83e45e0384fa6df83da79bd0b2c16ce1f693e04f23376a10a3c68935840",2912]
+["b9d331746b7e26d0d7179a68f0972c8c9dcc0a29c390fa2bcff1a83daca474ee",1480]
+["8b7c935382a8dd86e6c3816c9b0fdd4846b2ce586e6290826f0f4fc19a426c0e",3920]
+["a377e45bf3116609719ce950d02649cc386c1559d44a5a9ea1d465d92fb7a620",1520]
+["fa47faeb96239836ec5c2778b6f6db415accf7b6b44e1d8b332b19101c676320",4380]
+["6a30632f2afa0f00581246acd7da8665b1398da727d88cdcd43d404cd7f59ffe",1486]
+["13ea7fb506c8c4c16924264769fcfc34656c2d10165bf9bc6418348c4b0683a4",4052]
+["b8360de441ab18df9bac4c768528ad34c850a57d0e85e773d71e8d4e8bf0c5ed",1526]
+["0eff63b6750d6afd3b05ed96f7700ac97fecbcc0fbd6e2b0ad241bb04c2dfe82",4512]
+["9f0ae88b6288296cca75bb2effe8f57285cef5832cff5a9e232483e4e2f95dc5",1480]
+["05142928669a642ab5e6975574f16ce3faf913ed237b2b5c79c68e0a77e2be60",3920]
+["a803d337a4869a7f9e1f1dbf5b3256dbfd7111a431e61ff82c273e0a34f2a3a0",1486]
+["97711c725b167b9d1fae93e0a9058253e6e4805d2f39e93e6eff560b8ac7a2c2",4052]
+["195c898b8d39bec7d666928695029559b587773603686f73a61c586483b5e8fe",1526]
+["7d58a7043e79bb17fd5462ee7da41264a24bcce333f8cf525dd97799222993eb",4512]
+["bce8869f9c15f2b8de845c838ea4c9d5cf4772d77bff93385cea1fd99cee2e9c",1520]
+["977bbacbb0364ac1443b7bf2a8ef9a8dbb73a1314d575cb6c540a63a1f324865",4380]
+["8c45a92a79d9578560e1ab75cd0ebd087de8368afc3cdf811eec59ec6004c51d",1125]
+["422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76",2905]
+["54d59ea69049178ab17fafc1c7cabc7194e909d1d964c8d3ccca8b0613496c8f",1125]
+["75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab",2905]
+["9609d3cf78e5f1410e0e84304066da2a9c79b5570437a20042ed37f3e77d733e",1125]
+["277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560",2905]
+["57ef1854516eadb30c1e58a3b4ae561f8969ae08ae4bb14a4fb2914579658efb",1125]
+["fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba",2905]
+["e94df40a20283b19eb5080e0640b7374be2e3eff3a34b55b23ffaab163abf1f3",1480]
+["3f187226858f5d22a64cc117e66b29e9b3bcddd40b41d44a3ccefc751a24cbae",3920]
+["f1bb99352264637be542f28b8b9097028db6ae98550c4cff9f67cc326b03b819",1520]
+["8dabc1dc7f6fc959b5829e8f337d8bf3f9451bd3d20fbbf158028194e2e24cdb",4380]
+["58b6fa847337c6d5efd41c5e65162b8ff4252d583086b6fac224bb00ad1493f1",1486]
+["0c1db65f71b280d0992f5f7a02f0b26ec8f29a7940ad27ca4053f5f7706cdb63",4052]
+["b6e787b65aa3f69f884bae7b656a9e1030d120763f055b2efd9a7cab3fdf54ad",1526]
+["e1b8024cd9f9899b7506c35691697ad39ab8c87949dd50b1e4da17a66c9af106",4512]
+["cc3f7d63f1876e70a3c54419fd3f6e65a518fad880225181dc7f40aa4aaf61f4",1070]
+["d95ef70e36b56869c01fd38806ea47ecb36886eb63f9cbd4a9d0701b19d6c095",2571]
+["a9ac29ad159903f39b3089dee10d745e5b00ec1987fdf043289860bdbf7a97e3",1090]
+["e5dd5052f4adb5cf6285597ff5f8d98d56c52f4e9cc8a168e6474c82bbc45cfc",2881]
+["51989739a84d1f58755f143b1eb1b60e19f3aa107db44bd21c9dd113010825e8",1073]
+["7d0776a42f6f5ab4b217cb41f8729864e85644e423ef0cb2deb840102750a7f2",2661]
+["1aebec86cff2b7f78d6ef94339616bfacfef48db7553e9f528141b9c197771c7",1093]
+["b34672ad33fdf218b3c2655f989a778ed182cf12b80938672b5c0f6dfd0dc946",2971]
+["8095efe8b84dabbbde5f1b51f79dc63a70c3ab40e029e063456be0f1138b0b7d",1480]
+["b1185dbb32222d17cf326b89b28f1b9bb86f0a1ae1f301967d125fa30e72826e",3920]
+["c0cf91c057593594534be7bab0a9fdc90261ceb2396cc0363190cb1d761fc6fc",1486]
+["d6ae90101592a92b70f6a728041f133da1525d223b9c1cde4dd41ad108e0c2a3",4052]
+["7137c81602ca18c958c000dd8a90a0a1229e83513c61b0b5cf49994e79c3b250",1526]
+["780fc02373c9f27bf4f2a03a6120cdbbaeaa383d8fbfede59e7f7035b5e8feba",4512]
+["a70061a89437aa8e8e29c3b6ccdea65aede1d6cb4a595bef4236554943f2fbdd",1520]
+["82e325176d0c6158dd68e36c710dc47ce5a17108ac700aa49022ba27cb7770c9",4380]
+["22b2a64df200888301cc3ceb32e0fc0ebb2bad4ac2325f9883a8192f0c5c20f8",1070]
+["685496c56a4c43e91665ee5acded49c3c520d2055363e7336a429d16dd4751ce",2571]
+["72e1bdeff4a8686a5bf5fec86be5214dd7e71c43a74a2443bb0175d2c0373790",1073]
+["d3bb529b2b4b5df2b149bfcff52ae3114770bc975de6edc72c5a0143c5c0c6b6",2661]
+["eef7596f7c98918288ca81b1043317366ac969dd1df156ba4e883b3feda518dd",1093]
+["273864d8a063a3fc5c6ec37a3798c4cf23d2cf61fc657a2f3116bae03dd88ff8",2971]
+["26754e55c990784918e4434e42848b5f8c522c5b55762b5acd49bdfe99e7b2d5",1090]
+["6dbeb516fc23108c532fff1a279dc34b367f0742faee480c13b26f914bc79361",2881]
+["e317ce6616e6c4e25d623100f0ed2c4ff91c57ad1946f14a3276f0aba3e81564",1107]
+["d2df4639c6ee84a9c1dc24721fbdc434c2081e921bf20a768cec2189e862aad8",3130]
+["9515d2be53220ecf921ea4bea5a4c37b6131dac31e04ac82800e1c96930bc7e3",1127]
+["7bf4e8baba6a60d5fd829ae66cc5f19741d84bdfa022eac5d3b129fd8a216abe",3440]
+["1caa36635f6d19acec0fcc832410652e2f1ae2de8e7ebd68f305e470feef0e8b",1110]
+["8067c16fed3a472835b01fd596989d1899a0a72c8b63c70a6ff27c48c2a9ced5",3220]
+["20918dc2f688870b5c2bdbef37d3f5e3727b78b65583a1cc6d845cb46cc9cec6",1130]
+["5bec480d1e3863e30277bc2c5e2685d91de3088ca74fa13c60cf01b1259b417c",3530]
+["dfb881bd0e8292ceedd1d750936554481fd135bbdffe29d0859c3245f85253ab",1107]
+["051076ac5ef7da5a6ae441cebeec3edabd604ef3cbe5e9f5dfe69fafd019507c",3130]
+["b4c77e8bcd41b4f9d3e3207734a424fad353eb83a712701848beef2c190e7bb4",1110]
+["a8e4550403b398f2cfceca6c091608b2681c5503e847814c9acbb33589f98b4a",3220]
+["32499433742a34607e8b00263e27b2f48016d10c348e827d4a890efb3d7d7fb0",1130]
+["59840552af8e505e2fc6582070f31a13f9fa1a1f7cc72eba5b8ad8cef3e7c30b",3530]
+["62f93afb1930faeaefaf1d75e18c203d754267416a05bf7758a79b5c4bb643a3",1127]
+["088d97ac8d6c8a69dda6fa7ff3356404f90068625265ca8ebc2a24c454f4cd2e",3440]
+["6480e5a31c447b8b1155679c02547b49b6c27bbd66be2865ac4f2ecb900c7e1d",2268]
+["00ebceef3211016a92ea4d1a82816489442edac52837430b98e05fa8d0124002",7638]
+["3df6a6e3c5a9f0736431b60b892ecc2fa815c08186f801369c392389d8e55864",2824]
+["e54cdd2457dbe11a1b8045737a7f89d522d1b4faee3ab54ff1e15190f5e9787d",9510]
+["c0619ad44640591966ef01152c5506af27612ee55c4dc81e3d25d7cb7daa267f",2812]
+["8f57fa60e19b98bf9a700e8fd2da02d042bc586c18c4df0b6c3555189621eaac",9414]
+["b093b9d3eae1f246f5f6b40badfe3de2fef7ecdc83edc70a7338d21ff828f5f5",1842]
+["a18bd95502b807dbebb247341b1d7c7fdf1e7f3a6a267a884e0e4373c4069647",6236]
+["ed1153f3ec4fdabc883b9b4bdab82da9ecfea75d71501242343efc3a266ee925",1564]
+["7b2f8deb47f68495661185d367fc4442cb5be74b7dcdf0354b4b373db4f173f8",5276]
+["354b12580b2f772e5a0fa9dc7551f39f8f1bf385007f7dc6a6232fa8d93b8222",1842]
+["23ea253ce59f04fb95d50f76795b25c265644a105d36c8b01ab6500626784c2d",6236]
+["36d73d0e99c532ee78a03fe4d2f6ea2e83a1feffbd15e0a49385cd62d1fb770d",1836]
+["3da50352d76bdbf90a969932853641fec65848937b8023111732ba98c257a46d",6172]
+["bd72586c5a27a3af241e020aaf7948cec3ecc2bd522ec2f727787318df6dd2dc",1550]
+["c673d84c1a8a7dd2b3e3019404c65c5fffc83179554038b086ea8ddf04b51700",4704]
+["8c8651974661fd2330f9fd80ba477e6764022429d9e33097b1e49be3b65347c5",1590]
+["1290d27addf689f8e75406a7e265ce586a7bebe842e8293cff39a84dd158e7b2",5164]
+["cfd8adf1436ae36451c979062d19c17f7dfb0061ba488c515bd47e5202c50fdb",1556]
+["1223b13b2f1528b6cd2ed66e9f2e6646b6eef80f813e7fa5acef0a2e56044977",4836]
+["bbc5e879f87c23cb60fc9a264c83414195aabd3f49f58b7a0b0143cc4e858c4f",1596]
+["20b7dfbab2faf2a23a31af9cd9afe4789d47f976f706aaf99c52b12a094dc88b",5296]
+["444663a72518457ebce44a3bc3b1cb4545b7bcdf029295e153a646798e578634",1550]
+["1edee920f7ac54bac0371bee735fb7955865deeeec5d30df604c53d860e5bfa0",4704]
+["1aa12a606d286336d600de867311cb65ef70dd9c394365ecc689b3f853532b55",1556]
+["6df9e8fdbcc82942cd880d30d0918a13983c9b1aeab96832277ac72053739a63",4836]
+["4da09bdaa0e0bdf18d45e799c2aa2a7f2eba27b3057f56ea82be237551b73e94",1596]
+["006d92163fde18ed4272d0c9cd1ae2c8e3f45d1046f4b535f4915e02ede1f807",5296]
+["2a8bb494510bb9cefe6e0a0ebe3407c1cc6e3d6a447d38f6c8af13158fd1ccf9",1590]
+["19991a1020f4aeed661b90489d32ab158ba8fb4fce8c451c1da0bc46db179864",5164]
+["7d42469be44f7e423804525d07993dd7ce6a59845e8e2073b1b515dd960eef18",1049]
+["7ed9c6bb262faac9fd045a42ef878dd696e7088466afa21d2d9489017ee694c2",3272]
+["f10c6d4b6354fda2545cee5548a1b501261f1f787d68dcdbc371e91e036d714d",1049]
+["3f37501653da118cb5cc42657090cd526a5a7cc6aade930326dab2ec1d5b3e89",3272]
+["c0fefc5f16b3cdf271e88a2648b696cd93d115b16018e98455d1e7efb575861e",1049]
+["d654dfc60a8f88f16ce9680bf65771059a7da5eb4a64179521258085a33999cd",3272]
+["3dce63c612ed078b908ad0caa6448fbe4f6d2164fcb3b1249e1dda36386e108f",1049]
+["8e0a7605bc20c1d05167d1165d6dc4b75aa324651cab2b7eaebde84a7d1367eb",3272]
+["4b5b2438286f12ed303afd132a5f117beff7972a57f6ac3dd1442a1f156f86e3",1049]
+["7c9f17e5e29382a471b01b6b8988e8e7c43afe0824032197110094f3c17ef96a",3272]
+["1a80bf4ac7734dbf06170b40bd4439020413d7d4b8f41c39fc816248ee6e6d84",1049]
+["8a2dbf811bbd999dd01bf0e483f9955401002cfe32790ee6f37601fb6d713764",3272]
+["7772861551b588d20938d5e67a1f3e5cfa22a754fa331d9a57a1cd75b82b1038",1049]
+["0fbe8a7ac246478096f90ca16a45bea915025671c03d1fbaadd209bf4c12a648",3272]
+["90f0f8f9618f7cf03d8b7cd24cf6731e1f8a4ba4bf24e9ff33742a562ec9fd58",1049]
+["606c30f80917aae814fde13684d957d9eb92e04b317ee9d91d7a536a17eee909",3272]
+["479a543265e7f35cddff40de43243c82b034d8342b343a67da932acf716edeca",1049]
+["93262238ccaaa81e721f2eaad0859909b1924ec229f5d9b263f69eee3250934e",3272]
+["27c2a23e3e0167d0bf878aa39646f1e8f916e8865484b4b4604ad08d9b770f0c",1049]
+["d015c1a80c8ff4fc07e1d2e107cee849857f2c6479a49a0e8232a456c55daf94",3272]
+["5b5cff9399ad95b3c73e411baefb6ce0017d8cd56be183f71ccd8e8456eba11b",1049]
+["11e4ec0ba69a557796767109f35bf116a93cf68d9274e47f9fd9bce1fcde35e8",3272]
+["19e7fd94f1448071fa196c4bd02212d7e73ceb6c7e89afac0bb70ce6912f25b0",1049]
+["8fc7fa64b1f6ed70362867ffc541813d0b87150b1530e282312144b054703ea9",3272]
+["d1e613725183a28ee9dbe6c739c1b033d19c3b960cd17f70e185c87028520f65",1049]
+["01883c3ba3e103ef5887409a3b7435d183832cb06bd0697e34d997dc1678d61f",3272]
+["676a25a08cf90fd35817b0d3b1a8f90832bad7ee30ddbe642e3d3453fa575d8e",1049]
+["37edd9c4c91de177752d0f4e3f3538f0252e03a0bfa3b321b2891eff2a3813aa",3272]
+["8d4a9d2ea321a097cdcecab84e133c73b97eefa164119e9d3f396bcdad19942c",1049]
+["444e5a32824fb2d5b9d548ec0b380880922aeec7558bd169765d0793af994041",3272]
+["b9ff2f138f93875880782602a98935c98c8510d968c498e1e43d2a743e7b9577",1049]
+["06ae73414a5dbcaadbee286a7cfd2b920ea63a15dfb40b18f72e2975b57da4b0",3272]
+["57f5e012039a31276e86feed7220434f2ebc3a84e381788837778672e42731cf",1049]
+["60a94dc92d093c7deaf34ec9a710e00d1be236b40333cadb0340bcf567807de2",3272]
+["1ff4e222a9ba1b032b6470590f32af4b2e8ddc3c02c9255021d6f1630743126c",1049]
+["debcaaee76a0d91d909b36d764455865ba3548ccb57f1e8b4cb8f94d4f6876aa",3272]
+["96c06cc3eca92c51ab70767dad004d7d15366c4f688a53d6c57ea957db111f92",1049]
+["4e87f359815b80ef89ec307a237e3a32893eb8f2b8454408ff21c69ee998147d",3272]
+["b2426ae380536cd623228197c8f5b66f9a05a22fab30e93366cf687967321cce",1049]
+["e9a758ad6c841b2b5c9070f412b32696d54e0b14a431a61dcaafb7fd52cbc9ca",3272]
+["1c3dc89af00d74670a55b6802e4032492d267a119ad71c31c7502451e189054c",1049]
+["ee36ca93dc8f1ec9611a50a0e04be4a7ba14602889d0c09148895ca567c3ebc9",3272]
+["dfc84f1a12bc07e0bf83224ba21c5142be3aad56555b50a1dee9da90ca584abe",1049]
+["46e6250c7dcec9d80539009ee35c60c95beef69767a1f7c9b34d1510a3ee385b",3272]
+["3640954c25b20771c4eae2972a521feaaed472ee5701e16ff93dd9d93a7a9a8a",1049]
+["d754e968f5093a9231735bac37622b03d96d1fd09b8f059ca91778b501ec280b",3272]
+["e65ab94612620276995862dcf659c708e5700e334e5387db654a5ba2b617a93c",1049]
+["a8a7fad2caffda774b73757119226bd83741f03deedb2a0fd78d594014f428a0",3272]
+["577d681701c901550743d5b22496241302838f1573b373b2072be38ffea45b33",1049]
+["e252d8b741c5e48937009a1f05bdf66560fb6971072c62e6a8e36bf178b34079",3272]
+["ea980ebf8ae6b2286e1e5a837a05955f23cc6feb026b403ee5d71afe2088cd42",1049]
+["a84d70d10ab1849d69f723f6429a5d85525ed92c3faad2a1346cbadbe71a0374",3272]
+["01215e768069608fe8f4b6bffcdbe9b1d83936aba3d1376a4e9f15cc552b39a0",1049]
+["509a3bf15f03e0b00befe54edc3e0fe113c9751b6470a6a151b333cbf37f3948",3272]
+["866556abdabcc04b8f7a7924be47809999dacb7c98bb15b385be16be2e8cfae7",1049]
+["b7c104722f87d1e3c4fcabef98ec128dd04ef051c48e015579d60ba7e1aec666",3272]
+["7698462c388e032d1006af829b6a56c79af401d005c487f41b9d521f2533b2c3",1049]
+["89fdc2a8e37cf7fb95d3274cc8145f3717811ad6238e4676e61dfe171c2082e1",3272]
+["530e437771c598f9824338d7e3792b5fcc41de98cc5c31f02231060b446cd2ba",1049]
+["8c13a708cdce7c475b1a0cf3d57a415129bf20bd9ac97ef19ae778cf6224e3ed",3272]
+["8047ad33f72f6e815cea30e18769083405a7b9fe7087f06341fe31ac8f3fabdc",1049]
+["5929650a5df7492154286677a2b692cbe547d4c58b1c4afee36ae802ce63b9b0",3272]
+["d297e9356c7ede61bec7ce5481195f684021e001ac78183f4a01eedb85de72be",1049]
+["85088e74411106bbf19434db10f19babadb61de9351e3932c2e4f8a1732c78c1",3272]
+["60b55bfdb7e0ae3d7021de85a13c4689d88ac060f215faacf314f056e455dc0c",1049]
+["bd242040d10b33f146077de8e7f9adadfee49a430f31c32013dda90983ca57af",3272]
+["7dd529d960e415241a2a6f329027f21e350fa0f24e7054fcab61175f422cf5fe",1049]
+["9b09ab95d1f07ee113c163beb412a1ed283afe046599212f48be3fff2c4fd754",3272]
+["0e824dd1c38062e9b008337391a7999a7327bc7d184a1c20842ac3c37df99512",1049]
+["b2790dcedb72425c042223b9ea8949c82c29c7996a043a87d9854f730ba8cb13",3272]
+["76506bb6bb8b010e617263bdf06cacb7df850c79870d60ee7b68b0039a6b2e28",1049]
+["f6dc70ce6273d60afc709afdeaefc24018990b0da4a3db8482602920ad3ff64c",3272]
+["7c49ce1a3ca81a8efb1b4a4f287bd60c75cccbc1722b9c94a7d3080d75a7c146",1049]
+["06174735cfeb205287a9ab2cee040e56709906c3db65d77dfa7c020957259879",3272]
+["c9174f063d9eee0aca2758fb74526298240b96ccdc9b6c5bc9d59a6a679a4a05",1049]
+["8366d8f4c9e42b0ea394d839eacc7314b576e3eee356a3f136c3bfdf05c24589",3272]
+["2558e43428d26f482056976fc9c6103e5c8c8ec00e6a239dc1ed55ff621db719",1049]
+["819aac6fb57798e7dd72a2aef68a18686033f5ee404e028b9cedd3a602cf4d07",3272]
+["f9df92ad9785b8ed0ac02d804005d5018175eadbe949973cc1f0872809d24d83",1049]
+["106ea2bce0cbfb96ca1f77fafec5cb39ee156dcca50899a1fc9649417156b653",3272]
+["8f84903aee19bba3c7dad48f03e69b980fd437bbf66d9948187be378ab6ce6b2",1049]
+["45240d0b61ea493886422bffddf31c059c7f868b940fe574b86401a5c52711db",3272]
+["dfdd00eff6f8073048d8fcde57fa37d516fc57b8743764632e6d23ae7990c184",1049]
+["a84fe9c6112e98604573f24d1edfbde2c2c7827f37caa97e65bad9166dfb9e44",3272]
+["5646ec14fe17cf153290b8ccc3c739ef0c8e846a6f386aebca2148b49fe1832e",1049]
+["e906f2e6242afc889a9a02a37a41c794aa0f589e426098cc674d417822982234",3272]
+["8b51e5b56aeb19245aa9bce987113a74e8e5f5dc7fffa1c09d7bba7b23a4220a",1049]
+["d2934553fcc74124f04f85e2fde4ab83fe46456d0a551ddc6de5938142137138",3272]
+["560c3875a3071b5859a3ece026cfe586d324ede1351841aa06990b39d6828fe0",1049]
+["b7a6b136af28edd0956793f736d885226299a586ed4b765aebfec5d7f0525162",3272]
+["d7b8497bc0263db21fd4d3ff5006ef49f1a0e4006a928ec564a3b6f3f7ded55d",1049]
+["d4c6114e32e3fa21fac0730bc510229ba113c348c28d8bcb15701c257a73a6ff",3272]
+["31369ed739d221e997dda1d34fe0952fa36a5533e49c962593b2eeb0231a135e",1049]
+["acc2a94589d666dd96d7484ac4c1288b275d17d616208cad5bf27e7b5f7e0a6f",3272]
+["42ad42eac3951bc1e5af4a4a3d574f26ef986d612efca30ef29a9232a2d699d4",1049]
+["c13f2d5feb65716703d8249386e1118408daaa508eca8f22fb4cd66922844d15",3272]
+["00583e43f17500a81e9e4fcc55a205b30d013a0c0291031f202ae1d493ddfce9",1049]
+["87f596c52fc05cfea063922796cb68ba625dbfb3e5d1d27e5f231f33f3911df4",3272]
+["7508812a91c00a5bd891f39a5e2686c0eea26eadbe988b6c2364475f67c13162",1049]
+["ef688157543a3f92ea49fb0c9c903e6422b2bb5d2c3cee99d998e560a43d98d8",3272]
+["f7cfbb527246edc3b6b8fe00efd86912ccc81a053fe9dd0950f56ccce3b2178f",1049]
+["538234562b1e19e561f59a2540816342e948777936ffb9eb05822f4f84c3cc6b",3272]
+["40617c248a2c03b920a6a7ac2887f2dd375dec766c3f4654a2ff66087c1f4368",1049]
+["f20cded7713e27c53b2825fb37cffcad375a1ea7c642f4c5251558caa543c473",3272]
+["a983f381004ac33edff52d26d5feaa636fd21836a96387a208f649e97468a347",1049]
+["4492ea1eb66fc7fe2509e19808aa2ebb6b5f90d7d0b7801187c5073c14b45c4f",3272]
+["6b1b5df0bc837fe3075e17b67584af5ad228fd4f428c70906dc140d6ee8f69fb",1049]
+["c87df39fbfdc10923897efdf004bda7a5f553fa1674accdd8c5b778604b3f4c5",3272]
+["f8a6b7bb441fd9db66475c1040925f51836b95f37311fafe4190442ba3ff1925",1049]
+["4e209a081fb8552babadd5093e34377bf080d8eb65b6fe19b8354fffe6bc42f3",3272]
+["3b5308affc61b81741bee13d59f2dcb94e9b45063154baae8584524a25d0bab2",1049]
+["b98aa4140d4d63063561a7edf62b1002810589125a26367cbcca0e3bdb8bd80c",3272]
+["afcef91fdce08bca4b2e4d06ac84319c35625ccb65fd7384a8fef0e32a6a2926",1049]
+["528ee87f9fdff70be6f0773e3ef5efa8bcc46f12fc0640038f5dc4d61e4a0bcb",3272]
+["0663a331c65200592e6eb7367b8738628e1075b0712341d6b6f86ba4f7256e9e",1049]
+["22411f1fcdb33a0bae502e4a5dd7a1d515c80ab323b6e9da609cbac0f6cd6669",3272]
+["25a9d2357ee8bf5d6d1467cfd97b5eba05cc7e424216f9c6108ea83ebd803273",1049]
+["96cdad5475e5857b9959dfcb3aa942103086ab0a1dcdeb9017ce07129db585b8",3272]
+["8d2659b4e98a7b7e794ed8b3013bd1e45452ddeb9c084073e318e79a65223b77",1049]
+["a1d61b5dc7f72bde5da8708692acc458c81eb4ef4a5500daa08a3f87a8e2314b",3272]
+["e57766e40228e38eb104025282baa4407131b9a370d1039db4e0a6df7a829273",1049]
+["e7617bf2f2f56fd70a69e051e004473acbc1b3e3348a4584d3a2fb3d5ae95b06",3272]
+["ce9cf83b2d3b40fdff4b50c91126b2e64dabeabc4c36662c20b533538c6f77e8",1049]
+["5cf002774f6001925e23d1bb536239daa45bb500f792ebef0e76b4aef7007df1",3272]
+["ca1568e099215a85eba5d7b053fa10eb76ce5b58dcee874e135a8653fd942ec8",1049]
+["e1803061aa93e082e7facebfd2d376d061ad4e7bcfabf228008d4345f9e1e87b",3272]
+["2df297b811113425e87b35ce5b0cc62927f8a7ae406a290f28cc7617b7ba4d11",1049]
+["c214b13deefebdb4cf6f563d782fff19be0b3cb6210b0d25adeb204ce7954a50",3272]
+["efc884fd394e05df32a25eca9611356efc51ee505aa56d3359b40889868e8209",1049]
+["9033aa4ae5ba6bdbf6fc75b05f814238d7e560f2597b5888a547971ddbe7d0de",3272]
+["0270111cf036643e8084b18da1626377b2c2c4762852eb7bc6e3845e40bb4e65",1049]
+["0613b19df6856aabf191b879a62c59fb79dccd77cdc637d5303f39b013ccdd99",3272]
+["006933e0db618f061a0c51cd23af81e997a514709b052f73a1bf1ecb4c0380b2",1049]
+["b1326f70755105ee923355f252e1898c8b2ba227da35a03362a82ec6d5526d6e",3272]
+["7ca3f184f2d3d8828480d3341ada3f70995e85c80094cfb0659580f8c67667fc",1049]
+["857ed8eb63ba7304682d49123ac93f23fc2a26e554b64d33383b86a80af5f208",3272]
+["ab9b8ff0ff19d1f0ff15b2c50aef625affece77507d79a4752d7201b79fda843",1049]
+["756ca6106af1e64fef524f6cb6c3fd347af64a1bff2bd4de391a95a5c99f763c",3272]
+["365588a4ffec74e183dfc00bab147054530b02d9c2bc12be269d719a281e75dd",1049]
+["47e0390c250c0b7b8e65d5fcd3ab6cfdb63db34067772eae607f7f8e15ccb69c",3272]
+["0e7bf32f6cda9ece343eb5212cf4262ae5174ca527c685c10149838eb169066d",1049]
+["5902d4319a01f61c5fd9c1396b2b8c8ec881a49d89699f9f232db0111e1e4134",3272]
+["e7f7439916fc88c884ba3f0b919243d34582746051657107151b88e8804d685c",1049]
+["70e5eb5f5c38eb89d2b6d7b5e4125c4dbfdf70a523fab8dd715d0b21aa2d4c96",3272]
+["607ae63f521f75f7b2de7231cdc0e67cb40904119a8d3bd7ddd7c2649f61d6e1",1049]
+["6dc179738c8fda8f090bec3565f51f21d15f59dd7d4b770df2ed384c5523d17d",3272]
+["24454a336788d1d46f96d706a75be8c373c44a2b4f95c93aa3ba4ccf55acc252",1049]
+["03ed75d00b0f12045f779543d65951fc56deb5a6148c2cb5f4f25ec8f6c5f94f",3272]
+["7f6741adf623c832cb75f040f3fb3523db755152f364f96d5a05fa50648756b4",1049]
+["65fe5feebceecaa345fb6b812922170b974d357723ff652c07b18560d35c46f1",3272]
+["347e7a2710205fae324d8b20dc67fcf0ef502e640dab989b2145b24fb8c0a566",1049]
+["25663a3d995d433ef4c5bc5a7e65397abcd3bfe02487a3bd7ef954bbf4291c50",3272]
+["07f8ee47cdcf705d6fc65f8bbd01fe9a731d7e913dce86c4ac3a574fbf4a4d8a",1049]
+["4b19325f72174900e0b794aeb9a1726730e5ecde1162e9539d4f4651b0a31c0d",3272]
+["9aab8aea960a761c9b0a403c394cee411df7115a51dfc301260670b69c608924",1049]
+["9133c82813586e119ebd4f1eb9c8a0e98e7f37bcffd0c20ecf65c05d7ead0473",3272]
+["2868281a0658fedc8017ac0445d1b111fa893aafa3820c02456bdc5248f0a50b",1049]
+["0efa6d2732b93ca9ddbb4faa396bfa179848d373ea595043bfde4968aa77bda7",3272]
+["339c25074bcde90deba84b0380a600074093b325e630da0680a8fe094783b9b2",1049]
+["fc8415e7965c7753dd6119e9b5fd214fe4d838bdc0a67bb25d045d293e1d6135",3272]
+["e17c00c477c6f827c0e1ed9cad9a1299f5ba684d53a4f5a8740daa1fe0d70025",1049]
+["958aa865bef3a4992f216d9f756168ca0789900bc964ddf4cece197c448a9671",3272]
+["8053723fc12cba912b9725a6db5027c3c7871f8df5dcdd20180f224d9d839d9d",1049]
+["629d2df4100f52a7e40557227dfe1fa6692230d6896444e9ca12d65e3e397fdd",3272]
+["b7781bfb13c867acbd6f472105330f6cf06998d3c84dc9456525a4f521a88360",1049]
+["26061ad052754878a1340e1052369ff4acf5fddd0ecee2ec388767361db22ff9",3272]
+["33cd2e2ca507019fd91601c41378a97aa1995f5dbb0a18e887f1e6ae78acd496",1049]
+["d920a8df0e7929e23aff87d043237caba405e7069cf8900e4e6c3f677216163f",3272]
+["92ae05f3dcf861af2736d0631d64a6720266d52e658bae2d0760dd466afb8355",1049]
+["7947641bbcfbc4b6889736061b562cfd643f4f8f7119fa244919d4e221985be8",3272]
+["be4f792924650f4023e50326480da6ba3558082c5371086675eaff87975406e1",1049]
+["64e7d2fd20056a83e1c69cdabf92d87b14ea91bf8d2c662a1a7f2844df58bb75",3272]
+["4c729a5bb35d77bdf3ce7abe523f1ae217f64c7add1f76505072e77f8f3a2b96",1049]
+["d30f42e800a172f61ee22f6be3931b6f071eb36eb705b709eba6472a69af90bf",3272]
+["f8023d867406fee25f05d9786f7e10a8940896bfe327c885ed0fdf9dc64605a0",1049]
+["903299e871793a088b9a6b525304ff004079b9bec03220d70066ed5f7e88a3dd",3272]
+["79c5633d39e9d92c6c973984e09bde83b21be495da90561d0c8f125fc8d6eae7",1049]
+["dee9f403dd9af112d4c076ae31b8e99a2b57be6e999963e3a95e22328c56b680",3272]
+["a351bf01f2ef1cd1cd116c6920cd2c0e3a2a82044005e753ef25a87b448d3983",1049]
+["8e9cae825b39198c32d780a3501600b3149c78c2ccd69b2a1cd94da6468166e4",3272]
+["381ae975740b2f4e859054af826a4e48274ce29297209080c25f6a1c12b17fc4",1049]
+["11bbfdb8e336d1cd1ea1d0fa9881dc128e8d9b695dd9d877dc4382cd524b3cfc",3272]
+["248c69b24d9d040b65518143d38940d19172e898f60de24ce0ec97f20ba08746",1049]
+["da08b47dfb8186f9ea39262a176bc0ab96c0c300e4ef456cac77024e57de8fa9",3272]
+["00dc442371796bc3d49476ab5cefdcdf9faecf410ad18a196fbf4adca69b76ab",1049]
+["a2601bbd98ed2bf4fb0faece18ce2f84c9ca435c9fe6aa0e977570f4849c7c80",3272]
+["12636fb57691bef58982e016ce65b45f184a592f93c1eb83d709839358989d58",1049]
+["3305ce57099111a747f20b27d7832c151c67b0f40130e82b927d2f18b9c9f6d4",3272]
+["065734b3b7e6f0844b3bf803b5dd7a5b3bda6fb883e00b079ccd875759649bc7",1049]
+["e30e485851e9a0c3133eab61e9330ef2d00d36e7dc59a82450d86c78185f137d",3272]
+["5229a1c60fb4ddd9a00f9c11b30b8a72c6425ea6287b5e8a180a49a16edbc16d",1049]
+["7220bec8992f56e04ca99014e169aefe2f364c27177e47515a179611c3705f55",3272]
+["58f1d351f4d47cdc3901aeb584b881e85cf602be97fe97cc4200ff1dd65711f7",1049]
+["4643a7c978edcfa4a583466210dc3587bce7371d8dcbb4660e705a0a1c933359",3272]
+["ed585c81f575553631edb92e5dacc5fbddccc9793dca79496e96a0699f7ad895",1049]
+["78ec9542f9f4fbc13f16de815c1b8a4ba0913a557e69b113c772ec38efbca1ca",3272]
+["9ae350ee186fc79b64101c58281ab57c4007fc1cc0ed15b330a83d54458fb861",1049]
+["d883a938dc5068781b4d4c35bfb2c929107f73c71df6d96e05d696480d371aed",3272]
+["efa84fc3b7ff87f8875cc46e4b63336a92eccb3746b84d67d885656d8a560d4b",1049]
+["772a4070f57af37e43e28816776e360bea70070c4d8444760937c10a1bb9a1a8",3272]
+["dbc237d9fc1684116a3f7a6d735d5720f7e3273927e35a206154e45a79f2adb0",1185]
+["b0c8b12012a2a77585b4fb784eaf65e8b33840ced92db15231cb2c654fba44fe",3808]
+["a183b82401f71b912d87971917e90c8fe652601260822e902ecdaa4e7ba6899b",1185]
+["ff0c3e56f5c972b5c85d3663690d4217682770bbedbe3bac0b742cb4fbade015",3808]
+["00eb2efd5799c43b34e66f8b84f825cebdb3666fc971491fcbe75489f44ea594",1185]
+["8b9fa03b61e33d4ab5d79214ed137329fe79b830a13a5a0637b03a4e9b271d4b",3808]
+["31a283693269353690bacf3438675877eb5712452bd914a2fe67104d462d1335",1185]
+["574b026886a7738cec489bcd936ac4c52c54aa5b54d0b11334b8c1b3b837d71d",3808]
+["6e7f3525311356954c66ba5fac55ab3ba31d69d30f60b945347d90177f198aa9",1185]
+["39ac243db5789dddc51af0419d876de8626b51e1f31df2a0d289789c32507e80",3808]
+["e6d0c29d3d9dd583cea8a34a5906ce635de4a9dfa43cad3830e9f7ea2490637c",1185]
+["dda04f014a248ebe76bf1ea2de052f86f404ab5c32a166eea7ff259ad12bde83",3808]
+["b4e9442e25836040511c460004a483b26e4a6359a3fb8e07351b81a04286eec9",1185]
+["5c1473046874c3cafa1c84b5ad3bc9ed538cf9695e4b8912940e419cb39da235",3808]
+["d6a50e68ec3820e474c3893fd54860ae78d1e67543b5e95805630a5a531e8ee3",1185]
+["f4d07cd7d1678bed7b27e35ff8fd975106063812c813e8b9d4bceb9cb5664443",3808]
+["3755c2f2bd087e4ab33b6e1939789490919c3e060f769c1a0abbf0aef66e9f10",1185]
+["c779aed3fcd47e5bfef465dbf826b0e1f300100bc85cd28ee9348c35050a8810",3808]
+["fe7ded342652cbe8ca5796bc354e2c0c2df09af88df1968232146fc0ecb28a7f",1185]
+["d3084b41b07ff8e65750b1787db75a0ab0365d566b2fa53be4549f350d1af563",3808]
+["dfa5f43ec7e5ca4b4d935fcf4dd2fdec56d271ae4d243d9baaa3d755e33ae0b6",1185]
+["4a36457f7cebce8e8975f4cd282be892f3c0c19cac669c48860d3be0d8d881bf",3808]
+["b009de8c344a64ff1c3f09c3bca67b5650c13a11fdff29ac689fe81ff6870351",1185]
+["f76afae1d1bf3815f4472218048da67747fb23480c560a9c5fdfd5bffc5c6462",3808]
+["520e4c3e21cba54a1518e3c8d269f182dfcea3339e26ef870d2683028405cd9b",1185]
+["845eed3ddce00249fd516ebdb8e52b58b81aa742dc1e60b7574f16c9d1437d9f",3808]
+["e33402888bc3368410cbcc40b28247f889fe3213754b033bc4cbdfe6154acac6",1185]
+["9ff9f6a150a03dbd929057c6f24e2aacf89ff9c93361c5abb895bf8e9521fff1",3808]
+["ba94f34c1e866eb8180b3c2caa2be66093a4470a0cd0f4ff8bd57cd173872dd6",1185]
+["93650885615cf2575db0135597a701723eb674612676a6ac071e25adfac65af2",3808]
+["619e28353fc3614b6937e4114feea5d0df889104e0c79583fa245b8a6708842e",1185]
+["e5b9ac5c07ceb230b7328479020ef7182584d2df2eccbc8efa9d3fef1a95bb13",3808]
+["c41f9cc3b88ca920e1b5cd7e00fe9a57de35ad26441e01aff57ca188241aa32a",1185]
+["840f6d680fcc4427a741c23d02f44e74d725b9aabe227c871d29ae1e86e58030",3808]
+["f31577b1ef659f961109295f0aa48ba78c49bf3ada58580f3a491e07bd103f90",1185]
+["6276579af5d41efdd014b300e11dd27aa08f77b0757c5f29d1a2f9130486b214",3808]
+["c93805fd0dc55506ad3fb108ad0bdb5620e9cd52de12cb95589bcce10fa4230a",1185]
+["b7b79f682a2d3e552871360ce6fa4eb0c7e8c3ae1d67f1fc316da5c666315d7e",3808]
+["e7f2bbe0d6f9d91728b53154cbf2f577495463debf5bafa4e0bb73eea092efe3",1185]
+["a33289f237c3ebb023ee7bffff1da5cef24fb3d6bd6758178d3ae105bddf948e",3808]
+["4c469509c96edaabd51ed2207c39945cc8c8683294621611aa4a25c60b07d033",1185]
+["9c592dd30aa264f728af7cc065797557cff09f9d1ae098ae4ddea6a248f6de25",3808]
+["3524b76733b602c96c5f58cca3f2e8979238301129d03c0eee33e5e0a2a5956a",1185]
+["2701e9e264f3d00278a111b64b57d5ba87b9b3c493e1ba6aac819dc90df3c667",3808]
+["58bdfd96f163560d58025f62a2baa967cd2dba220093b419c670871828c04053",1185]
+["80afa0ce94b17731c93702ed7530588842fdbdbca2306c24d88cf614f01a02d5",3808]
+["8dd0dd05275c3184b0ede32735360820fb582596f12def73bd5ce1d17f0ceed9",1185]
+["424be39d8bfc64c7c980cfae3eb2041a8b0ab3c6572f83598e9b8f1c40584534",3808]
+["5ef0af59a7be3edc53ecdbf150294630d03af32d8c9f90ea5ad601f148789d68",1185]
+["76711dba30d88042d0e8af829e075c6f90158351264f7cebe480a65dc468dd89",3808]
+["fe8269ec9eb1161ec9038d9a7792e61ffa61c3866af12bb0a14117c2ce29234b",1550]
+["a6967e5da26a84053b1ae4c1f239a52be0ec0f06d22df009e0d6f166d2108c6a",4704]
+["b06b4834fc6906a63caa8924a0eb4da5d47b7fe8636a6f2870cfd0f5bcac0214",1590]
+["ae20b93f5c7b5993866d165dd989ffd74481f19a51e60f4b5b73e4bcda1cb845",5164]
+["a2d81eb5138b6232d86a572e597f383587b7f71a512e8b7bcc21f72f851ac4f4",1556]
+["058a277b934ad10a0d02d016b1ec668a661988df70fb769c362319fb0ca372cf",4836]
+["de66242fa8675fe252ad123c10368a9220b5dcb52c20440725576b44a3c98bfa",1596]
+["0da15d9c40a86b733abd0eeaf2069d23b8b0528da629818f518de2eb7b82bc8c",5296]
+["b31db498fd126a8a69513c41a20c1e7735d61e1cf9f88227a1e45f40393cd44d",1105]
+["1a35faded78d11a074b43eac6b21a757629fafab0db10b18b88d0139c86a6f49",3099]
+["714f8130b07cf6de950e828fe433210e540f64135f9f07b1461ca5a4b343a8d9",1125]
+["8d11474141f2e3111b427a4a8fe126606f6c77e39d056bce7b07228514b8fc9a",3409]
+["00428f0f012f42efccb02e88759f1faaaf949633ebdb949b69a64dfab44d6ee0",1108]
+["a228956a3c125f09462490d9ffac5ffd5130396e85918e22c46813bf637330a8",3189]
+["200c1030e3ab513c76b2a5f0e3de34cf37081b24fb1c6f656e6c02bbfcd54b4c",1128]
+["3ce3430f997395bcd2ad9f5b30f0fac3de00c0a2f9275f08ee0a71c05d2327b8",3499]
+["586c513196ed7cba01b022afc598b23e348e16521100ab7374f90ada00a5f051",1550]
+["8f8f22e61a21e9a0e1878c03a3f43d0b2e09db2e80ad8ecc414c649ffcad6f31",4704]
+["0fd52eeb4174d3ded5e6c1c87b2540f0c592f1808962e28351e73ce6734a5a83",1556]
+["5e8cf697147b987dc138e613faed0271996429934b479a0426ce5e0f5fbafab9",4836]
+["29e976868ff9e6519230098d64c3a4754d26bef1f13d6331b4d1df427a326ff3",1596]
+["4bc73a0a00f81683a940dae4c7dce36e5d8121d1aee4387398b0d90ea1613bea",5296]
+["9181d714df50a1450db6f697880237968af8edb52bc1a40ebebc093f6dd1e67c",1590]
+["90cee4bf7cd43f5a32f9659c11f9f1093431a89d637c4ab63af8dde6f2e9d006",5164]
+["5f85486ec529d8d8bc77b55bcbe4816d2d8e3a65b1c7c746637c41167921b9ea",1105]
+["5b79f20c28258a186ced56443073c6ecfb33bfd88890d54a5ae9e943d32f947e",3099]
+["7a121b4b44216f76670eebbab757e71f46265918314bac9789f2c22fe70595d0",1108]
+["f93e5c5c5cec24f5339b8fae8b04d4f2c924939a96d94052f5e1942e02eb3a8c",3189]
+["6face721f63025d5e2cdf3a12e32fd0ba21c60fa9a5c7fc4934b53501054d870",1128]
+["11cc67ec4e4516bc9a7a53bcf2074e660f15f61dd58377c2a44f24a4652f94d0",3499]
+["05bff5a8c2b03e40b977815ad944989b529734c0cbdbd3b425ee1da0a4f0301d",1125]
+["e23932b72389ab532ec6096257cd9aa629ba608ea4f53c2402f97b0d7a3b8062",3409]
+["c0375ba81907548e4b96f4cb12d03bcfc825dcac8dcdf947003865702628814b",2816]
+["6af1f888e41c5ec6d8b80c5cd57eb9a9e4eced889b44c984df953f8ad004b83c",9418]
+["6a0ddbd2b3d2d4f58cb2cf67e8dfbabc4bf9858a709f16121b6cc0f5e7005784",631]
+["f1e1b7d8ad879abd120d6cb32a5999925af0ed77b3a7144a7dfd8db76ea18246",642]
+["458e51f920ec49d95e1d54ce239b9bf7394e3489e21d4ffb5000f4c3bc77570f",586]
+["41efbbf8c7c802d5cc666a6b8a5d092e31dfb9302d5d60e03f0b1b7ea6fe3772",663]
+["3617d5732c341946ba6b3e75e227aee240dfb719d8f19689b5f3cc8d47613619",584]
+["9e300e67a1ab1b8161113429dda72b5bc5ee125bf394ce76d3dcad4644948531",655]
+["fa1a1ce660981782f39153d14b405509b21c870a1b519d75c12ebf526254b65a",584]
+["deb2ed1abfb270b172ae078504892c74782a9fee9878f08d32c7aa5b5e19eb57",584]
+["862ded318ec482ecb3975636f4ac8420887a43644cef569099d6cb43b84c3612",663]
+["2c8d8644173cd7a1cdd7e4cca28ae73b2d9089445c9ecf696a9ca102d425bc6f",592]
+["2f9393ec865475ee2a9070b063623b1c4dbfb4cbdbf2d5ac3bb1fe6630eafd06",695]
+["bf0f3bd741bd6866b5c6d4b54f5c0f3826242f69dc57f41a650ac6baeedaf481",627]
+["2d9426c297f8b89e42a5385508e945c764295b28898b6b67acd4f8058103f0fc",725]
+["765ee30650ca46011f93cf07cfff7166023dc6fcd22b9b635b8e6090c9c282be",621]
+["87a37997f80f59e247f51631bbc35e1dec95564e024a3f3ed2262836186ff71f",2344]
+["2706fa2a3e6eb15cd9f61eabeab63ee139f32ee2dd3e3ed44801193ae188bae9",637]
+["1ef6703039a9fd293d150f488b2af1641c7b76cd84c6b29f8ce36e4a87b3ffd4",735]
+["68a5c9c44f6070de2f0f91f598ca0725145465ca0541535fe670debcfa3a65e1",628]
+["ad7fcb907ca47eea31c3b7c609c6e04eaf5c4a27fae085fb3a3a158215398179",2344]
+["3e329ac41f9d418225039e6b80d50a9f4236e0645ac400b9600c2711ef75ac3c",639]
+["e18ceb1adeec2e3f83eca18746a8197dcc4ee571d32afdcf8db75b109db1a474",749]
+["c7d4ca7a397824e6253e4c5d5bbb6713c787b817c6f39ab15f8c2dad2e64d690",653]
+["0dbe405c72e2b9491d1bf4534301a10a4a613a402b70a0c6153109e794ce9767",2360]
+["62f74b7ad72ea6dd8f710cb1c12acbd64883a587090a44f3b5bb58a2efe4915c",627]
+["25ef771a6a9f188a1c7a281af6728f2cd85182f7ebbe7a21f5a547c3036c1212",725]
+["b54cadf12a5532404536e148f87d702f5b75791c6e94f6df8792e7e79bab1995",621]
+["7725107c710b0d2f55ded61370cef0d4e890d78b12d3d22f873c089c432296af",2344]
+["70d5cd84c8f1f88c04c4d8447b16216d249786a2146ba7cbe5f5a66879f3712a",637]
+["e8ce2fe9b4531597917b632681975440999db3943ef76b5140a61a2f6b66d83d",735]
+["8d548304a5cd4a8ac658dcf186b1127efe980e66303bcdc60079dad7a5e331a6",628]
+["5b6bc599e5dd7b6ee02106f969165ca6b04484682a1e14771b363c0c98fd9517",2344]
+["64403ed43a0da6159fe4c58e7617a225dcbd132de0cb5efae798b7770b589b1d",639]
+["bac2e1a26841b99ac022921537bb214193afd2dd32cf63c4a94091056d7df771",749]
+["6b80dd0c0a055183c643364268fde9b4939a6685ed46e26af5b3b3bd59362a51",653]
+["ecd327058bf16ee2f1bff10d375abfd9b96a1e93a04b2ec4e909fefea93ae199",2360]
+["a6de543266faa5c6ad8f58a4eec07886e149affa437b889c91649becf15ddc0c",627]
+["fddf2e2af800462dd5493f0d09bf98fd38a44cc863a34e1e76092ce001ddf8fa",725]
+["966714bfd358bd9a1a6735ce8b1e4db623f46edb730201093b555637dfd87a12",621]
+["3d84f755f2375378d074afc59ba7f55771cc69f86bc43efa79e92dc25cf201b3",2344]
+["a146b24da9bfc77fedbf1e76ca818a556f6c1494f6c8eff3f6262f14f2015b7b",637]
+["166a9587a8fb0b495fe08542a82dddfea53f4813911a98d1bc34fbc8df821085",735]
+["2dcf42c9fc085ffff3bdddd727d751f87a513e7dc221cde3b3608fa65f916ef9",628]
+["f722d33720d51dc5b0f9b01bd1b2100226bb98fe66b002633b6c65f252381455",2344]
+["588b75dc503d23e39536c4e58e7665bbc0d97c0fff3c91abf72efb9e05cdc3b4",639]
+["29e1ef056ac874593ed03bcddf88ca1a9e70f0a80dc67bbdc6b2b687676e6639",749]
+["ea9af5a7ae2cd9d6900306bc59ddd6871f0515a60a0469265bfb3ae001f48c5a",653]
+["f300a89ff0e9b8ab90f64364b627e4bc34dd5d42206f2904efc560ea67f2f25d",2360]
+["910d4e8d95b4457c2ffae1e398fe53cab1dea6e35879618ed7f57ddf7fc86a97",627]
+["f86d200fc72f7ae6c16e88767a5a28b713373c4f26628890eca35beaa0d54e31",725]
+["3195d395f484663f5cbe3a7531702a3f18fa6d513b702791d80631c8621b62c9",621]
+["1f9a588fdd143e01c413e8d851d5065e2a9b6a2e3d609cb9cf9ec9e0b860cbe1",2344]
+["d600b3d047e7a6264bf6dce970f0355b41a528cb25d191a2cb51a980b3b684aa",637]
+["7888af7598cb85ba23c5989f89a0aff6ca45664f411801139041dac38470ce17",735]
+["8680e6440f359daae51bc46903703dd06dea28c77b74007e5b4e9bd7e0019317",628]
+["02e2dc26a35485475bbc01d60acb14dae89a63c46e877c3bd7261c7d109b8072",2344]
+["607aa90aee6c2402d05a7ef8df9dac02ae551d2ea309d74749a7f2bfddea54b8",639]
+["eb8fd6c5709fb66c66cba103018f8e1fffe5b134b1671e02504a821b235bb497",749]
+["c2a03324e0a71d8fb5b65a3d9a6c3a7b17dd0b959e9bf895822b1e9df2275553",653]
+["63f13cc80f0c23e510bbe6102e0b5c5084ad5f0521153252e74026240fcdf2f7",2360]
+["f850aff7ffdab762756a6d51af7d54144314ec12846db2f914e6901a25abfd67",590]
+["60e9f43f430b36ba232cfdafade6a6ad66f14d3a5da4225b1e866771e8d60ac4",497]
+["d9cc9a63aa49b0b9b5b936805a1275962cadb47020c4acf7fa2af36a91bb0998",590]
+["0471bad2179ac6612a4dc9f79c25d1c9f27b2a9442085803eb00ff22aa3162c1",497]
+["2c7bb88a87b2cc2ef1e64de68926b20bcb6263757307689b4434e0a2cbd45c0b",555]
+["b2577898e3a6ea631ef294de2be64abce0d7a0bf713fb109b979883d61ac6095",568]
+["cea372a56668529a6cbef6dc0d365f5ee3923973780e7fd32df334f5e25a162e",553]
+["4a524d36c449d628cab1794fe4a5ac61bcd9c83de8cd0f8cbb3fef6c5a628edc",562]
+["7f355048e11f98bb8c32506bc5c6f2aef9d16cf98c405305f3083ea8cfe12c1d",553]
+["9e781c9bb463005162df2cfa86e96a4952b33ffb2dd9a1dba916bbfe2fec35fb",587]
+["e75c32dac80482b1f246f261afc2adfa5ddbfa881c1a59ebbff23b63966c6cd8",634]
+["dc9c104c371ed79fb43572d06f69a54f4ab0cc9142db1033a17d867246520bce",460]
+["61f5c0940d8ad56b26318eb6887dbdbb9043db9cd50fc3e5b753b7542a028bff",2344]
+["6bcc0d6e1ae9393fbfe0c6015485fcc06d38a56fdcec0393778e3a656e628226",593]
+["4b4c65c96522ff846cc33b4e78f740252d9722277b9d3ea7d3a0f63cb83f4eee",646]
+["f389f603c0ea7f6ddae19fbc0871f7ad1fc40e332403ceaeb4fb0fffcedb7276",469]
+["4065967eb3ef8d8b1a87387fc2991ed2a7c3599e9c23c9249a8ab09bcda0df1b",2352]
+["5e3180428f970e843c8ec8f7441dc988a74455e8e19f8b920b27895f711dfa75",642]
+["807d0cab4237fbd00f97cd527e22457836e36fd2697c6da24ee91beaedc98674",807]
+["498e47540903e04e7ea2164ee87fa776cd3df9ebf251e68645134e5067c59bbe",640]
+["86872ee441867d6978bfc582bf1bdf3092db5ffff1f0970e1f86af9cda6272e4",798]
+["442d337cf1ddcd759af1eb20973e772ed6861d14257f07de82430a9d196d9c1d",640]
+["c1cc8f9814891419af30130dac4770e0809e05125a7b6ffd704e7d1d0f2ea955",519]
+["7e4648ba5989cb0c08d58c0aa1b791ce7cae5f01b8fca4409fd764724fa1627c",381]
+["e349f08232702f92e1409b48ca76151de9f6986b45fbdff6a863bb4a85b207c5",519]
+["58b293d3fe532faeb4a81026138e757ded575bacfc0316e966ef1b5fdce787cb",617]
+["03e197797aee7a2eaddf22d6ea86ec03e16b5ed7276bbf27c0ad87c2033f0296",2360]
+["5533b17bc7d62b04e675a5fa3da2173eb67e1e7555dfdf290fd6d9ad7dc0f96f",687]
+["6d68bd0a0409a2b0d594b19ac1798c67d73c8e2817dce183dc423d9ddba55695",544]
+["5ef6e19178da7d10a18808420150e7e2e7c67a0fa91c5b09114207bbcf6d2f4f",620]
+["ddb032fd36cc215e58b8fa4f4fd152a6cc85c7b9694f50dd869c4494d2baa784",2308]
+["86274f5688125897157226a6ca5e7d721fc76c0016ebb005a7c37f2d7a4307f5",672]
+["b6d735966c869092f8dd1c5a0415a9e7a3f7804d73832b8d362a0fb92e25f35f",657]
+["8bcf49bb553c0fbfe895a0e60659f94b6c4d293f3db045fbf1c90cf2fcd15bcc",620]
+["3fd244fddecb92038c4b10c11c3fb4cca976d76380f1428a43d4b3a06a9ba0e4",700]
+["4682f7ed2ee0f260530b9902ac1d7fd186e35a7da69517de77ca8a21b74859dc",2336]
+["76735e93f676689a52bfd6cecaf65bda0c73938c5467fab5c4b80b59a3590d9c",632]
+["d3a69fdb715a43a635197eef22e6b0b6af8b7ebe62b91a28097b8678d56aaf6a",2324]
+["cf346fa70922aacb48efd7d120c8418d08eeda95f8ef5af80929db43709f2171",692]
+["5cf764862a2e0c596435988cef37e52c8b5d5d41f9722c7acb2c32032e98d179",697]
+["4d4ba4c03b13875d95fa56362bb032548ad3041e72a166ab35c68ea03b892d31",632]
+["7dbdeffb07366639d6211c220fb596a9b3e681e98ce8ec94ce542df8ff547f30",724]
+["b41cd901060482af86d9cd686fc76092bd9607bb4227cbb7e36d13ef930d14ac",2352]
+["3c144cb58dcf044a6a03a2e83913213edc0c08c46aa6d750dd82c8a6ff34a7eb",605]
+["34da7872a9c24e4b681b02081a65088d6b88f5e2c25ce7928a2108586bdf382d",2312]
+["ab27910d6182c7eddcdaada6fc243e830bba616c0927ca2f002b038838a33b33",665]
+["ff90eeb4bc1e87456c4d132df22141cbe8b41b64652b2c7e7b53dec1003d68b0",612]
+["acf31fcfa16faf8256ce9cdc58329e573be2479e74d0ff871b9834ee5a9e5e21",605]
+["e6e88f2aa4b43b87db0344660fc574c69c24b4cb48f36f233a4a822bfe1a7e53",697]
+["9ad87eac91aabd07484167d3a3acc02a3c2fc6f7247e48cde15bdad3152f6e39",2340]
+["ca56cb1acffa63cc5edfd6eaf5845860aa263533c7af759888006fa90ab02e27",605]
+["21d68d4bfa715cbeb478b938e29dfcd017591162f561ee318c73f27a725e7ae4",2312]
+["6bf5a8fafc32ec99fda27c6dd9c1f1a6bfe886d5b5792574e2ffd025704f6693",665]
+["d31f2be46092ca29ead03a7d5a230b0e1362040b573d27ed8599bba91a94fd61",612]
+["faa9c1a405f79e924977b1c11a6ec17ebc046c85f23b5812ba9cc1cf1869e4c2",605]
+["9086e31c947ff475fa073fcf44316a045ac63883fc4786e77ed4ffb44c5d61c1",697]
+["d906fa3666bb868baa631a8a4828d85b0ec399589d4a34c17cce8c77dfa389bb",2340]
+["5b5a3d6cc974db060931c57e980282bdff1e3f8ba7cef52a9f4a1926af7d7b33",632]
+["e2364d21f7982c6aa320d4f52ff06b3ec34abde7aa4bd1b347dcd396af749f71",2324]
+["80f3a6e4864e4cc3c3e69dc803702c8ab570093295d4b9875ca135eff4b71174",692]
+["52f34f4567d86e0f57dea86188a0580bb5e2694a58f7eaa356c17bd31fa1ce9c",697]
+["b395852dd1f24dc2f20685e81aa601f732e57286e1ad3ffb6f9c32da0e02bbd6",632]
+["3f5a0dd7821d62bffb53828fb3c0ed3b3b02c788000140b8ff288d51a0d1dcfb",724]
+["9b364444ae0b72aadead899e26f06c1f00f0f7f47fdb2ca388f38fa55899ad1e",2352]
+["0a9247467fe3cd3924494fe00b5f9e200f29bf1f41fe8859ec5aa1c1633caf88",608]
+["aaaf03112d53de317bd87516037bfbcd7e8e1ea2c606c5105f3a96d9208da6d3",2316]
+["8f7a936918a8a3c6e93b83a4e9612309fd1de88f98780cb1268516941df61821",670]
+["1f9b83ceaf4bcf6dafd994713326985d13765126882ff38dff7c9dd9b9b01366",620]
+["41647cbff32f02ac5b654ca50c376827fc2f6c95389878269beb786fc2395bf2",608]
+["8c7db5575b89cafeae4a17d06c166ab6eabe2af3be5f4ffb63dc8652b6b8f56d",703]
+["2c2da0e2205216796fbcd95afe9cd924f46e498060d6731078aad67074d14d9b",2344]
+["4a7942dcd7383eeb53e4da5a5c30190c8d3ddb470f5156e1dc949b60bd697d2c",605]
+["b85dba46a27268bef2eeda90ce93da29fbb05ff26d08da24912017b7535c941e",2312]
+["27b158586e2cd1c8eef9758402f5da7287c8cf3a3e189c6762849ccc5eeab989",665]
+["7aaa689f9ea0490ad33b446038b400dcba8f203160cffcb1817e93f69eede7b4",612]
+["b946255fd959b4a6e843cac3830dff11b9bb5d466eee0abfe65b876259aca2b2",605]
+["a84975e63891dac69f2f47926605c0d2f66f0667e2d1f841a83ecd319943acfe",697]
+["59d12e32763419d30fdcab75f93e039233946e385cbee7149a858ecd3201553e",2340]
+["aba543ceefe44bce83bbe591b02fd816adca61115be6b45b51f9935a443144cc",618]
+["e4221d1da0c3fbea82f1ea259170d7cea33547400e0733fdd6a627108d7f99db",714]
+["de4892784c42984c831e5a8c76dea5671d8e18714071eb22005358156bb7cf33",578]
+["24134d8f2af6484f967af96a5a344fc88d2359d8e698017dc04114dc6c93d387",2372]
+["a97200e0b6cdc4bfbd55ffc888d4d9ce47beffd93d737f449ceb03c5e4059af6",625]
+["a5cd99a97b9799db80ce1b61d9a7d88ae9b97c4e0541c3cbafdc706a2f9d84f6",721]
+["2b0f18121e535b7759bec9825e662823843efc91f9b2d39b09015eb893615052",583]
+["0ab7d7acfad4f399797eef34588cf277967ae49d36290ec0519ce50b8f47ed36",2372]
+["d6813bdec8cb276cdd23e72b2af1e4c9458933966ea0d429b577ab4b68e46ad2",594]
+["c7b06382a6b0061232c0462ec8472ef76569c1df47cca8dbd29f72a05be1e272",666]
+["64198ed4b91f4ab98dff4548e83bf69d382a1ee095faac2092dda81db0c2aa1b",529]
+["57819bb946551581e0d6c3a0c37927f7f08a9809b590e3ad9dd1ceebab3ca47b",2340]
+["871fbe3835d0c89dd692a9be440a0716cd94417c7549060b15b4ab6c61bf3087",601]
+["af6fc1d4e21de5219a5583c71f15d7a9184e84933e0e08db7a0dcd5106725daa",673]
+["40a0b6150d1883ec00bbadfd5e4432f50a6cfc09b4ae16bbc99a937ada211fa2",534]
+["30562117b90af5bea32c29c500e59577bb47f875dd11d7e279aa1bbee6e4e72c",2340]
+["2b7af88988f519afcc57f5ba937c65b13c83e974571fb0d4f33f536db09ad20a",606]
+["bdaf6c12bead523a3f552b9b6211af96630a3696fae807db8cde0c3f808c3fcb",690]
+["fa978c045c2f3d4c3d33fd5b239bde34a7d1d1364d195bfd4756213a1bc6bef2",553]
+["f82eb65375d0513da58a12c316abb9297f32f3826eb0359b9ebe23aa28585ca8",2356]
+["5dc2548a709f7925cc182ea4d2ea160b67b9ebf2b3a1f34dac7abe921caf7e40",594]
+["ce7e4cd0d2b1f669a8d87e1ab939bbe9f9aa2135fd554405af53335753bb5fbb",666]
+["e1095ab6cc961f90906ca977f7439756ccbec558bc521fe3de0e64b9d24b5d27",529]
+["37ad77885d3f27d42a3bd418281dbcb0050c844c55a8e6d4b79a8ca76bb341ab",2340]
+["629400e149ee42cb3f9f36a9e0a3770adc4e2bd7fed405019f388a3c637ec2d6",601]
+["16aeb1c502e9304cd334a3e992669007b7c5a9c29f39b592ea5423618adb8d3a",673]
+["20313e55a51cb1d2d3c7819469961265c2a2fcc92115e653bfb94c3f3d628d69",534]
+["d1af5e3ec9f9092c137e61c90ccf365e428bcec3385ee8ce2dfca6fa3b4104a4",2340]
+["dcd72a5d9069b9e186dd6bb4cf34a596a8e7f8b215623806f13e13e0eb60e39d",606]
+["1016adde061260397c20c9d63b36bd2311c3ff0dd05c52ffb0dcf0ddaa846e34",690]
+["2a525eb3a32f0230f279776dd4178f06e72d0204905fa5f450de6aa2c0126ae3",553]
+["32f56d14af013e70a3c593e85bc4c8b2375efcfbe075685d55859541ee55e636",2356]
+["70446bc9021a9dd9c8d8a91ad47c56cbc23c0a64b8c4914fb0434f18f7772336",594]
+["dcadfac88c5abb5d12c20c8f447e06555e93782bf5e1a881cca4126982dd1d16",666]
+["7c8bce5f3dd89b3e7c69c17614ace523748895d5cf1d4d4418845f1b85b0b5a7",529]
+["7d24b25e66cc5eb250f2c0016bc75a2a3e12c3069285d430b3f4495cc098f915",2340]
+["6dac31c4c9b254fab717c0bd4141ab42d38f4691c9ab6aa553ae1fde3f4d284a",601]
+["8b03b6ced96be78609924ad70499e08df2ffb9428b04d83f6d81db80c4eb66bc",673]
+["d2f85eb479c85e6ad31ebe62b4ed02c5df94663430c6d8bd915b7f897bf4afa5",534]
+["a2ef13c26a0d41438825c7329ecbb840845da7d46934f5b11a6898e3b1dda5ef",2340]
+["94d329be5a7f0050c194d62919afd433c1c2d8a3d933fabd5fdb312c31e71dc6",606]
+["2841db3f5f497770a92dfe86ccc23778b6b6cc7f3c67fefabd1f52da16cdc466",690]
+["c21af2f1cf9768175c2a0af056c29bbcdf7bdc1e572e8b90519753391f1b90f3",553]
+["830d424978348bf99173e62f89a30ca3ca65420ba4387518503ca2736db51b2f",2356]
+["593f274748d78f1e2e0d2cc1f86f65b5ee03f683a17160a0d2a69817725ed687",594]
+["532405dc7c8bb31b9af3a100b6c899d574ac0dab64fb4e5bb2e5413305716c6b",666]
+["d739eef16e2d64ea69d5441134ed64ced8745a45c35af377d7e636d89d070a5e",529]
+["512b98ba48b27347e006b5f10cf8a641d62c6679d678c6547ef894c2e15383bb",2340]
+["ff7de75629dc55fdb63dce1ae0922fcff054a55c5985c34c38703de3f7b12d50",601]
+["3d8f7407c9d2eac2be005a6effda4f230467bc9949fe0f2c4eb75096dea86a0a",673]
+["36f37dc19246f1a85914547133668392ded513a51582ab87a28404cc324bb4f1",534]
+["dad86b9d439d5ec916f511043d4849e9eba481f8ff858e1ed1a85b56815fc550",2340]
+["602c496937b2ad2dd8dfc4b762d273aab68a6422021f54ab12273b9fecdff280",606]
+["fda6566c8d63e22f165e3b0773df8ac809081edba0b8c94bd5c2916ed4b19cfd",690]
+["726ba25439e1971e162405842ab5f3bb6da7261ae31baaa069f6ff72b309dae1",553]
+["2485be5a42e5728827737da09e42a0ebdaf96f20e1398bacbc833c53aa29881f",2356]
+["acbe548288b4b057a29f4ce771e54a5cceeb1e7acc8faf8cb744c485e6417ec3",639]
+["d5fae9a19de1b249238ebc5f92d4183c9732dcc2f610e0aea09e2b3af365ae09",2332]
+["17058f542867f36974e304a17743bac7245cfd63284faef3879bb723b9360054",657]
+["f45deee145b7b97d46b3c8f963f4facba41a55ca3cc3e95efa078d0a281b1286",657]
+["0b31d8cbddd0330560ad539baea74762361cf32bf577a98f09365f5365a672cc",639]
+["1c1fa14e304f39dc823ac3d8f985f0cab33fcd80abf9f74515cbc209af1037e1",2332]
+["2c2af575243e1705d25c49bd41db704acbfbc5509166304b93a79ead403403fe",657]
+["b68b96b424e29590471215aa4e315485e7f65c157c78667a93a69bd31a679ba5",657]
+["fcd58c2c1ea7eee4f9cc9f9c929b23839f06d59f10b83e6dccc82b88ccee845a",671]
+["392b04e61fbabf6085fd91b9156c7bd73862f01298d8bf1fa9bf1e97da97f30a",2368]
+["759b419d0e48313e98f1cc2ddbe239af0029902578db25c3fa6dad6d3fec4a9f",705]
+["5593e4e0ba02d49c75ccb14caf7ec53a27dc41bccee4f47ad1273f43a962b963",724]
+["3d38c3715d303093252419b59665f0fd8b00f83f89149f620c4301e63711b4a0",683]
+["7b89617abd0e8040a85b4af0241a28940a67ebf92c768ff0ca40f25eddc6c610",2388]
+["de78a84d94fabdc0550e7f9d7091fd17f205e24350c0ef406d9ad7e9a93c82d7",725]
+["41ab3a1c22bf191f180861d616b4046b823cee1bcef40bf7e2a818169b851247",760]
+["8839592af8a83b5910ab810dc0cb7b6203805ca2eaa11996349f42793986f2a4",659]
+["c9808c480237e45a70638149d39072ad76d80f09078c28b758f922ffc6c7df1d",2352]
+["6d511bc570450116d89e59af39ed0c92503822471fd0d133f53feea57e8b5ff6",685]
+["f3555df59e49f7644d9b795467f31d325ddebcdaf72d53c2018cb92ddc6ced4c",688]
+["5370174ee0146fb50a2c19cd3618bec7907f1bcc747737eccfb3018042e2588f",643]
+["304ba20321833ac99b7724bee304df929eaeefd5d6b687ceffffad4a3e27f90c",2344]
+["09ea019784f2de7856684a9a6441237b0e0383ced82761a42f298e1fbf5f8d27",669]
+["4488cea67cf9b264cbc7a57a62b8878f627d09df52ea49bef641a025746f7e9e",686]
+["98c5a6fc2d44e055c9a039f48046204260a2e32d4b3ab0beaec159621cdbe92f",671]
+["b34dd35782e48bedeb7cf1d7eab41cb4c6e849240d9c936b598b4736c41e9bae",2368]
+["edcdb2518063013484d48e84bae66c5ea584d1dc15bb4fe666fd5ab7a25cb3bf",705]
+["6d70accc536dfebfd4f18e69c341ae1b5a7d33786f3919d14c0104fc9dba248c",724]
+["feebe38f3fa6ea73f694d2806ea391f7cad98e18e62d3f5f752536866281a02a",645]
+["96d631f765aa606402f0b396080da3e89b0f99b6530e8c90e1394661a8688db4",2344]
+["ea4a48dada55331275924deabb7e50838b697581bfc97c9d54a20f544419eeed",671]
+["0cfa13d06f9afcaecf20ab13dbe4b9569b0d5ad30d495dfc35f4832c19148551",686]
+["f0c6923f17cb41b31ceb8018dc0908125ff33dc00c07538cf1799a885512f6eb",671]
+["a9ce4b7d89575eff0ebcd8b8e2f900bf65e16a847eaab9f504d9019b23106b2f",2368]
+["513f304715f941c43814876ec1b14143561377410479babc0c62878700687d56",705]
+["b5518211cc8709667508e6022295acf70a452329a83edc78b2296f437dc6893d",724]
+["2401822239f576f0d255f6be320db40fec25815e34062fd9d42541ad4f1da250",683]
+["0441448af7975029fcfe176cfa1b27e45255ddf7c9c4f403208ac4fa1a222ae1",2388]
+["8a422631eab6c4aa05295ac45b07abecd9023dea1bb82a6ea9f923d4d36843cb",725]
+["fe65e09a79627db8110579b7275020de6a9d332704278ed6e90236c6e35bca17",760]
+["1309d24296f2de87ffff0b72421bf4ca2d03dfa6bdfea0b2e24325690979c8e8",659]
+["499b8e484ef873a6567e418337092a0446cb3fbc3d7ec8cbd37744476b137199",2352]
+["a31510a419b0e3632d0828589d6ae31d61cc8c4cd5b3ffc0d26835ae55fa01d5",685]
+["d67c2cabfc7e780a06c97dbb61d8a33328771411f0f12293e963289a5b2051fa",688]
+["56fc0ba15d8542e490317be6ad95b0e969cdc8c474ff708a0442d21397b082a1",643]
+["1781439bb0e62bcae48eb2f213158eadc794ea771cde8afa1b5b07b5ea9cabb2",2344]
+["1184d3ad4d1e876b91c742713f7a1faf31d8b8c48934c854c1d5b11ca782a3c3",669]
+["12d399989bac1e17950c8f223785d103a1a93bc1d0b16cab682421165647e4da",686]
+["dc40b4839d860b4d86b9d37bea3f6a87717399c8ea7ad24598bb59c1c98579f3",671]
+["59899583e7545c362a5f8481e9a13687b723c5994db4b9cfdc09dd658c6c16cc",2368]
+["c606f35dba52bf15f0a4269b968a0d83eb75b87bd35ecef66b60c7f8f069ea36",705]
+["5e30aa10805a26fc90ca10a4291edeef4a689c5ddc0f3a640d6d7fd472c24aaa",724]
+["dc1d5f327fa7174148b4f01fc909f92daa6cff29ad464baef0aca37c3efeed1a",645]
+["a2aa9345ad0c48644c7e910c7c9e5a8295c6bdd7720abdd48d4337418616772b",2344]
+["8d370a5c147b5351769606eb91943b7e426555c6d59a8e88fdd21f95137f3985",671]
+["ae672e9e9f559a60395777f6cdd663b74098fbd2a61149fc9eec6026b0e403cc",686]
+["9eb0385c98f9a59a60cf2b719d7921f825f1cde7684e7de6e15107fc2dd1fac8",636]
+["e0d269f6332d4e2eab8e21c8ce4e680953a19cf6906a3e7286d5412d8acb202a",2328]
+["dbb0655cd7f8dd20d517f31ff0e6b755e7eb16462e27f8df75ba9f73aeb1f58c",652]
+["310b151470bc8314dbc05f7ef12c00e5058343ca8b78ebb30dc9874c1372635f",649]
+["1f744a6738929c1b3d529b1ada8282def5557fee0c45b3e0cf981843d6cf1db7",639]
+["166461804e77fb46291b94297ddc38ac00d7f5af6a32a0d2191aca5eedd09754",2332]
+["d5345000c5a0c71013320ff6787444829345b5cb28e0ec07ab11fa517e4c1564",657]
+["4e3e72dced992772a22255d809126506ae4fc73119d68397cd015faacdf187fa",657]
+["8cd472d3dfed04f81f04c3650e1ac1688eddd01ab9d017320e0f17adf38a489d",577]
+["f593428c9ee9849518bb848c2b70ea0706426cc863d762a613663ef52c289429",604]
+["4d5bd971fbb98f527270d66dd843cac79d25bf8bad74471712c313636604393f",582]
+["7f048521ac95d97f95f83473b65e03e489b7d92a18988a1696b3c1954edc13dc",609]
+["a017e2a501091c47511137f3691dc95bbe37609682f882f33d6d26ecab49e9b6",585]
+["decda7874c613c8973b0dde7a430465196e4e72859b7dd7c98cd7da26ef0adaf",631]
+["696f80c00cad8292d5008c34df888aeb753ea085b579e054270a930d92d007d2",639]
+["1590347fae5d137cb5f577542eccd0156be416e081fbfaed44acfd92818682d8",2356]
+["2da1f5f12d913bd4e473eaab7eca9e1fc69815c31297443eff34364766de5ff8",691]
+["9758b60ad9ed87b7e0a90f9fd57c3edaa90c7c793621c5f911f0fcaef0a55555",388]
+["97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95",1720]
+["ae2031936303bf1870cc812d6b61884909002289e72e99e820a5148599523b2d",639]
+["810b8a59b514ac3b0370f9372639ab758f2812207232c43fef43d5230707cd71",2356]
+["042c859ccbe87e7714cad11f5c251b63c500d7e9cdced1f6cf215cceb1968539",691]
+["1212a8cf1cbba526030c0f67dbb854c5dd2d24e7cf7977ef2137a14638be0d8a",388]
+["074353212d4ba5b400117a3914e962cb2b61bbb89e239c1ff30ce20f276e7662",1273]
+["99f3407d1adfd806aa6d0b4b453110d0ab33e37706911e8a7f0ff4db0431acba",2792]
+["c46df21ff2541ddec329067f062fca6d7b00ab3ad8ab7a0c88b43a4948d51436",1308]
+["931e60ab62472e660c25f8f4eb30d817c7c57dc3195986b9cc8dd9d8e1f51527",2696]
+["20c3cb373c6a7b061f007b929d5b930f1adff06754b041fb696197cda3f14306",1273]
+["30e2d3ad4acb5f6fd706836881ee337d957570304fd457b9a2b6ee2b2a416190",1353]
+["02ae0bf8d0dcfdd1b9538430c96f231dab80cab5ba2f9dd7080fa426e151a04c",2820]
+["5af78184f7751c1fa5ed0c5e333effcc7cb68ab3414acb65d6aed8c1b004d28a",1213]
+["fae1578a13740dc8e44b0555bb3b29f7c1b5df413caef93c74c570639fc53de9",2764]
+["6b28a931a504163218a104c66c90a8350963ce4c271e899ccad0f04a3486f346",1240]
+["d358181758aadc8a8b20e3069330124796f8f569a807a4157987a54658588a9e",2568]
+["ea5105359305da79969cb519570510e0c3c5aa9eb121567de3514297157c9ba0",1213]
+["ddbf401ce42ac8262b18ea139ff2526fc8965b355d7537da5adeba18a58b0c3e",1281]
+["e77b319519e6d452a0f26d7131eb3787250363ec4e7b435c86d2b773740feb18",2792]
+["e805bc1843e590b7755bb1d8f80161298ba6be613fd94965498ac465e9d98fee",814]
+["b7313a3a76b824e9109c27bb9bb6de727ac4d7b7ee5f3c3080dfc6baf53ea9bb",1008]
+["cf9e63260666c97bbce9e934445fa385bd01806e56a1492b595953a18b122898",814]
+["4a5878ffa474242ebb66a5da48618acb92055cc9d12f6da298bd8b7afcae559a",1248]
+["5ff95cbbdbb4e5701d1a3dd2ff4443676c614dfb40e78e9cbc930d0adb5cc07a",2764]
+["566ab2d0959e84b026301119046f34fb191c9739afc19c24109c5d861366b867",1273]
+["18931105adc004c32226c713972457e19c4db56efe010257ecbb90ff9be3296b",2532]
+["5bcc0ef476bc03ae581ad0318c0df00926cbfa52ceffa7583701972ebb9ebf0c",1248]
+["8365502f00f403138419663b1766eb326355e7073c389e76f7618b0bcfe56618",1313]
+["bd2597ee1cc1a481241c05c510cbb6eb654692654abc4526539aa36deeb59366",2792]
+["c58251b437842db37c9dcda918c3f55bedd5d3578f87149f771f88c35e427ea5",1188]
+["d8be2c5790fe33213f858e69a0744c7d0c0d0219936099770cf205c5f03d4deb",2736]
+["13fca1eb07ee353b2af80fb1a7666ec3971df40797634c6b03f706bc96f3415c",1205]
+["42fc6632c13cf3ab9257a59d09cbc914f5e0749097f06f035a731ee885852ac8",2412]
+["9f278f45bc4f791251ea9cb12f0fdfa89e813d8c7bef1974a3924df133205241",1188]
+["551ed4f4b09977792fabc1be04213afdbc6190da6f5d8ea9834cc422b0a6cf41",1241]
+["bf5d10378523b014a1a3448c931a25bc1172675ddd0e5d52389cf3b5fa4db059",2764]
+["ca1f988d468fcf9f8b0de15f0d2333b0b08c721f9fa31d4b8d628cf8190b960d",807]
+["61edf12f4da156d3d387e4b47cbf9352cdff095c6a5df102af6552c852e4bda7",957]
+["7167b52bd4b3f92ab2c54dbe1714f64f128e213a0e1619d75317361bfa9a9ec3",807]
+["b0944273cfb586c7108b0c5f1da76352f7419b05c4cc0efba427aaeeda9c5bb4",1251]
+["23abb79cbefc07768b5a8395003f762c1cc30686fb522e0d467d55947ddb915c",2776]
+["0f9fa2f02c59c1b6238b08a034c402ea7d536a68f3df9b12298dd8701cb1e627",1282]
+["690fd3617a4f54cf6b3b837da833cd5e86b84b42953f109aa7f1179cef433662",2620]
+["56b608bc0e2659fd529db22725a5a777d529a8818a21e9aee089c9e36084ef65",1251]
+["c008cf96d950988338803a074283d577f971054085f546abe8b6ebb408c04f2a",1325]
+["9b51baa00aafbe7056fdef2574fee69e2a8cd473195ff564633e186cbd5c1d41",2808]
+["ff6d0d2750915c7c2b839ec8dece705dc95834c116a010e49cc8c6966c39e327",1191]
+["db4c117a10e762ec6ce714ca254c602ef5cb2844d383202d88d23f157c4d49fc",2752]
+["1fb8fa0da58100f1f887d41e047d4f2c7f5ddce2a4bf873744c88177742fe23f",1214]
+["42b39a12d65292d3041777a9f86c11259d679ff8dc00ca8ad13759143744e51d",2492]
+["f817a315a36c89a312cb360a751577b09c589a8477ed8124cf7dfdb4647fde04",1191]
+["866b2bc6260eaa1b9846eb9c19eb0eb91271e9290eb5ef7b2a565d0c3b5b07ec",1253]
+["03a9f068186585b2e0acae2b648fed51fed47a6bb8a38eaff70d84fa078d6d83",2780]
+["2b7ee78abd28400082f33d9ffbf0d6b5a3c426009963c2da42f8403b10d10eda",806]
+["62e7c8b6363ceeadb3b790d17404429feb3303d1d9908e56ac0a6ffe24f36aae",982]
+["b511b88d5d3faf3a09af5d495c5d6b79700c73fc8027d53a6483dcd6e3d4103f",806]
+["c631b9049f2f14bfb6253274cfaf783ce4921f0abf7a43643d40f7c5331c0c78",1269]
+["54955720f1c634c2394764a88f66a0f19fd9a91e1cb3b4f9514077d50b0ef127",2788]
+["f810f2d097d41a3dbe7c22be4a7ba3c6192fb783d3871df8cb6e7e274efbcbdd",1302]
+["d1ceecb47528d55e9e3ee467691ad3453699fc581682b1a6a74bea22341a1246",2664]
+["bf1c28a0f13ddb6d62673fc86b31242a17a6ae203636c795d6d6d0069d2f5451",1269]
+["1861209da77851fb8db89500ee6a9cbc7a5d92369cbec7d62298d0c41f23fe8f",1346]
+["0886446a3931238754aefdc01879b07ee1ada5a6ed26a41a76ea21ac7c8cd9a1",2816]
+["e0b71389ecc9e546069de2ac5cc6a9d64db67b801c2d83af803f0032cdd558fd",1209]
+["95c9f8ece1665e4f0906d18aa4db199117d84ac081012adec85d2c53adb45699",2760]
+["b0bff8fdf985b88a5041b07c6cbe03f2e8b5baabc55ab24caa6e13410b1d7a95",1234]
+["7e5d6f1260b64511fffaecd6c9c69ea7dcd4814afde93ab64c5cc587ab6c058c",2536]
+["f8f84a6c8a96c778f4d52232bcb05642d22a09ace177f379cf366e58fe7f8f4f",1209]
+["e24f90ed4a4369006dc3a80d8963326430b0bda8546180096f64b4a42d9e6532",1274]
+["32bc633e5315870669c114a80497fc334194fec54f31659d5ff9512c09f63c1a",2788]
+["5d6b42c2a61e94a540dcfba9c54aeff9b37255203f9bf7af69ea2dea9f352cee",813]
+["c53896fba1301476c1fc4e73b68040bd43d7adf0b4a589c870c70830d7ebfc64",998]
+["721a23fb88560521953f38f24d4d2bbd0f57230ff4c06cb2ba43f7a40a223c4e",813]
+["b1d2834ef0eda9826a0101adb2385d4acf56e6bbe2a00b96d196631ea792d454",573]
+["87155c6a0698b70a6cca99ef9202e36f8ba489e6095ad6794a113a4553f5d1f9",592]
+["b1352f059e7726f7a0bd769c04a2f0581a4769749bab3c9f5efe90338d538140",581]
+["63a7579dec309564be02d70ca9a54263771540d9220cbf533578ec1b0328c87a",617]
+["83a15f4c4640b2ed1aad90d78ad4a021c35a4866fb53fe5772f9740427995690",551]
+["38b4b1bc930ec099166f4c2d2c43b0f1508edacb60b19cff2a31c0abd6400a28",495]
+["02f268cca95d4533dc8def63ba2e71e75339df67cb3e7068061ce24cea0a1f1d",609]
+["d57b0a8473acd3a7f7301074697815c9a44b457df30cf4dc5f4f92777def021b",696]
+["f3bb730c2118870ea8f9c0598c4ba22747d1edbd83191e1b69c6d1615a5c1595",559]
+["8ab5d9aeb8fb76bf6086519106eecf1f94ca4cf1a3f1e020fe1e5e52c457e2d4",2364]
+["b1c73f14130c8d8a55ffe96a2c6d282e0599462253566dfeb9719ba5b11ca6f4",621]
+["49bb457a8725f7d232d87de91338ab70359b708bbd4ad27262b08202933d503f",720]
+["c71d052ab74aec6ee2aae8fc7dfadbec987139271f8c3a2a6f792f81c61c6451",585]
+["d3862185482115636d1e8a440d20389cff2c0d39c1bc724e8643f2a5fb15c54c",2376]
+["dc1a804dfb274c349931aca4ef29e7e6a6d5cefac487760f66b2a405c9a28730",597]
+["dc8b852ab3a5aecbe705fd9d4b6ed547015e3af3a9d3e4fb7dbc96278accda1a",672]
+["374a572645ceceb7394e5f944aa205b35c6f754d4f01c8538510602082d50ef5",535]
+["e7b6d7f88bb15be33951b76eb12dc1cd6a60105f50a3fa5609f3001e22a75d78",2348]
+["746e23881c8fa8e149347ce6ffbceff8aa4c726d03651ec4b429dc2c3a5060f8",604]
+["b9cf9554f2230a7a30905b0e1d079d3efc1319a8b2a7148ad85be9f50f53d7ca",679]
+["3bbb8972270c57655a33d76bf5633fff6e53fe5e92ce3ff7387c9edd931ce490",540]
+["265914fc501e0c79262e2bf9f541586472c5a6c92faf86c76486f2168ad0560f",2348]
+["832dd40d7ed04910c2e8c16e5af9304d9d00b08f0c38f1d4e8c4c847ecb08257",609]
+["bbd048eb55b9712a06d3efabcc64030cace7db449a90d24245236485018766a1",696]
+["02725b859ae3309ce392d4c9071e7871fd431c1707c0cd5c755b950ea9d6d9a6",559]
+["2c9218a2e7b1ee24c14bb8ff9e238cd0624801a6fc5f46a5135ab2c2444bb36d",2364]
+["8191c989457b57ec5d182798c13ce1b18096582ba329c4b29152f2c63496b73f",597]
+["11c59cd6011d296ad84c99f7cb5557b99e563f03b6eff795d2dcefdbb2d806d7",672]
+["843ce31d94dad09f355a3f9cc8752f072840d005e4eb488559b0fd6598286579",535]
+["6f7850a84d4d11108605509458ad97db385732c7146bbf72274b8d2a07febf00",2348]
+["bdb2d73304fef3977d6572e1fa73fe1f2575c42e2572aca6d97445ca098a255f",604]
+["974a1fa3b0588aa07a8c7b3f9a0a2283b29966f2c83629eee36b1af5eea191bd",679]
+["34024c7d10caf926ccec3d62bef6232473d4cf3a5a44be46e7cb10eed26008f6",540]
+["86fec12c31de0c62c54b766ace214b26ddcdd87ae28daa8fe524f34202915d10",2348]
+["163b0b602ad30739415f4c14d12a59516be941511a1908525861d3154bdc210a",609]
+["9131db1bc255ddf611a87dffde6bbf23184a852c3818c8babd19c4deb4330365",696]
+["de876f2392f6b25c2adf545209cbda5bdc233def990d9c66d464c3e2e9e61621",559]
+["4c09768e518b6d6696161fcb31b7115b53592a2495410f3b9341b977b579140b",2364]
+["17e9608db04922240a2e969fd0b59018a5beabf81c299550d249c9c3fcf7f0d2",597]
+["964a59ed57f2882dec93669d9e376c6a137cd872788f406838dd12d0df7a03b9",672]
+["5831a0f2afb7037f82f3168bf97e9b81242f7b4f6646ca90841b56984a931712",535]
+["296bdd03492167bcdabb32cec8623956aaf0ca903184e1b761e242fdd5cb040b",2348]
+["bf9d5d69853daf12f6f54abaa1ba3dbc589393c913ae43c29ce5bd2a884b29c7",604]
+["75f2b71d1c7750ec3d33566ffc7a3e5f713503af4828a6f614f0b9d167b3850c",679]
+["5a7c113f7bd46da86f4b7dbe283cf52e2c7e7227d198c5e81bb7b0939085f1a3",540]
+["74bab68d7aa62ff16e9c2444920ced4bd20c06fb14840a445e681fe867bd9abf",2348]
+["08fb510a5bdf895d8f0ef85b2c8188acc80c2c5eab74e85dfc337c2c37a200e3",609]
+["f43a77a9ea522fb8c36289e9d6f6449d3a2e88d0b912310ba4be45f1ad3c143d",696]
+["8e751ed43321236609a72d2ed01d7218f3dd06a4eca8898c55219e266e2028bc",559]
+["c7e1d84a6b4de5b76dcfb553318c733bb2fcb0d77845d148b4c836e60e5e79f7",2364]
+["87b8905ea816df3867b4bc1509256ba1bf4b8fb4c504f7912e81f11bf864a132",597]
+["e4fa1ebddcaa9b4e59bc1e0701fc3b42392c28e9a0153ace4b5741ecb7c23dc6",672]
+["726d4b06b2d0f3ccac01fcabe67ee5d98ae960cf0df1f8563e8f15f048b24df5",535]
+["7ff2edf38b3efe4dbc48d3dbc45f729bef6ca8b8630218ac61e8d39d0534f94d",2348]
+["9754908c469a5ccd00b77e017efc2decea90adcf15a5a35282f3d007b936b1f2",604]
+["4eaf920b824b8757db39c4254bd75b34f7d42c2b9493010f77a527d441f9d1be",679]
+["39ceb489dd1f3dd0ca20ebfe3df8f8bac3776c70accc8c800c9205ac20b06b70",540]
+["d44d25e1dbdac4673835a567501bea5b810e133dae695a0d92c17ab22be0c322",2348]
+["4f3abe90a7d6ca7673aba2a90ae7147455066c008bfd06b77685a0aebbb2a05f",609]
+["da701b4dea5f5e2a9f71a8623860e3721ed6b7017b2121d5d1d41a15e74cff75",696]
+["133a97059f88e02c2f252172afea3a6a960845bfdf5fb5f2834488fc7d05ec25",559]
+["c7069786738a93f1b30d2f414ed450f039a6982051832b03ca0c4ab870affbac",2364]
+["723cd73c792c472dd43910e7d8d5b66f41e889149bfde6f5e225a63068f9842d",768]
+["da39a8f6b4250cb2d24e73800ecf1895b26877174fe51446dc29b212d8462d5e",2392]
+["3e49ae7f2a122d57b06ebd6d7774d6dacbd38ef9961d68c68bfa3df9a34cfd6f",797]
+["d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64",825]
+["d3eed13ce0f84b1dcb12fa28854469259e1f4717f70a0aebcac7be1878610ca8",786]
+["2fb71272a31b7081cded64bddabfd82ba3b734816bd9188e8de7a82bd6610955",2420]
+["0b656b143112dd023505698c29392783cceeda779094e9acd4b05a1b13744038",827]
+["1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2",891]
+["fac1c3b96f6fdbbe2431d362c957c86f60ee9e4a61c155a6ab4614830e2f5ea8",814]
+["0d6355942930b85f9ea7c75d2a55cb7b64dbf333e872f66ec2be6ea8fd86a8f5",2400]
+["12e532ec5b39427709b4f48f0b93bab21cd3fb24be30bf99f4986bd9c9b24bb1",843]
+["29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6",1049]
+["d20e8e537f4e34ac4a2c5605e446e221e173ea3dc9709adc38f89558c1e41ca8",832]
+["f685f2fb68f8bb19ad88632c05a51a780c8e1269377857c6c73e3993390bb5b1",2428]
+["14d1d0eb6bfc50c2ecec72048bd1e61b7e2156c5f493a2cf845e671fcb13aeac",873]
+["bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1",1139]
+["4465bb1bb889fe637cd4b2384c707910a47c893eeb0964d860cc0931062f55b5",586]
+["f5a1c4f536ed1ac3abfdf512ae62d2a14d076e68239bd2570f126022c6a32781",663]
+["1863b7250b7d993dad5706a0a3c9b21efe9db894757cd325f285d350aa0e1b86",584]
+["145924c8a4ec74a37f4b1d0d24117a69d646f637e00aec05a699a1c9659866b3",655]
+["5c48ac5c57b0dec9941fbe3dfae96239b185e0bd3470d9207d2fa20c55db09cf",584]
+["8de151682be073714d2ccefca51d42647f6aafed5bfffe06fd5177d54bd771e8",584]
+["8168bde50453bfd49fad2d804032a9acd4685627c8278c8438973a5832a6f493",663]
+["5f9b4f7388eeb2e9647158e2ff73bedab67af53f2cbce62930f5c7ede2ab74b4",592]
+["e78a07579d00dd7c5f876e826dfbd29ad8f413f2100df114fd38f072a0f9aabf",695]
+["d41c7733601ffacb006d1dc8f6d7a9f3334cc589afdc523883881510d06faedf",690]
+["92a6e0307f1351efed0701d7afa68086c5c2a7990b0a67e8717e6fdfd46af690",2384]
+["18910e141ff4be0d120fefc24c602e25e095dbaf8ef803455e9404dc5bc84461",629]
+["404b73996b67401d94cc82eae5b8332b67b40e08df65bc906c9ea8cc160db84a",716]
+["4e58effba1f055af3a51b2ddb829761ebd205fc37d3f771b3ae9b735942667f1",690]
+["2d16f1ad6ac4174b6635405801eff15988b6536e8d8187afb9e3356de851c99d",662]
+["e5b14777c5d7b487f8d38f461f15de7782656a23c07a634b7117dd47859b715e",2412]
+["02867497b852bbdb49239f274a15c528306770953cc90308e29da92b6ed69656",595]
+["d5e7aacfe0abd2794c4bd2afae1307c08b2f35bc67f7d278ca844505f31f3ac3",2312]
+["faa66b35dd4bf0bb2a07c24a0a549a7ff3d4eea8368fb9703d5cbd3ddbe04f91",539]
+["91b5a9910f1b4c9a6960bc81e8a911639a26d156f06fd769a9d8c49db5bba627",357]
+["f6956ad2cf6c45324b2eba3758425daa41402981fb2aeacf0437b9c1610bbf60",595]
+["fd3aad32ad30bedf3889b03ec0e5a2937faf7d1dbaa9004c29498db421aff2cf",575]
+["242316d71c49cb44be1f225ffc4993d23f0c74e17f7993a67339432f9b1f0b1a",2344]
+["26bc9ebcf1db4b3c3bc3ac0040df30e0c72a9c590be42ffca800a8ba1188584b",575]
+["e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30",598]
+["4f10481da1d84fe56bece8a3e5b0eb49d6c6c4ab569ba3eb8aa12feb4f28a708",901]
+["58f2f7e53b5c0c8a65e29db7eddc493fae2838d66c86c928740ad28543f0142d",804]
+["cb91c04e40a91d484cde8a477874a0d342c72baf0f4a477a593175116e52f5cb",2408]
+["e51ebdbc07d42dc7c61c4dbb31f926a3ec9aba0c1867db714fc4ce55904df860",905]
+["6c1e2af4c86788aae417b1b55c91e66a290b99773778101525d2d2edf632ae08",832]
+["b301214a0c41825cbe117bac108bc538bbfd4c2fb8c02128b86c18278f278854",750]
+["360c33b1a0721fa17f8842dbff3c64b3088adbc8b7383a0604393e297b1abca2",566]
+["81c4479531f97d592615a8103984cc40556b5dc0f9cd7365c4a68b97ace1510f",754]
+["8f5598d950b3ee0cec951c2b40f01769442d9a87355a4e32d395747cb3751925",587]
+["5dfb6aee3c82b7c55dc2c5cb79641cee4a328b68a1e245c54cd2f6c3dd4f570c",828]
+["1f45ecb91d777432e0b78b09a3c32d847c1be85fc81552afd56946ebb5ea5734",800]
+["622c17db5cd70d142fbe63d973110cf44a120b73eb588cf974e46523ecb9e505",832]
+["195ccb73e6c06791653a2d23771bdd19fc632512f83a2a5ab3e69e1cb0f7e6ab",832]
+["2a57b1f996f1727dcb8badc1f53570fb9a67d1a7a3361871d591dd403b986316",783]
+["fea5aa4b4fc6f8778643e101328451eaec5e054d97bedeab3183c0c261720355",669]
+["9c58bed2dada213807f8f6039089b77be87b91114add83e9de4f5e740e74fa11",787]
+["a01f6f6af014f3ef22f3e77ac016c9092d2e48f157ada184e13508df66c00eb2",693]
+["56f4b2dfb9c4506adc048bb7ba33ef127bab7f1cd3f43e1d1cab68e1f9a6e484",762]
+["c5929254050b20d7d6df41a8c529adc5742820f7a8982fd6844665dfeb31fcdf",704]
+["0d153442d583c21a93a7f53b4ac633a7fd298494ce05de91638c3a689c5c5934",798]
+["d7dc8baacee5bac09bb9d38f81f84839ccefdea3c9ed4eed486132673da92b13",800]
+["06fd2516641fbebefed1033b983ef041b6b70671c79a72818d0af7b451313524",802]
+["7ae55b6c0d61fb4e54a257975b3ba92008e9905d5efa0d49b615a31c9538e984",832]
+["1a3c387f80d333f31924ce3da8a5aea40f3ea0f86b79fc6b8e2f7bc5f31442c9",798]
+["598af19690ca5170093476bcf96288160412c24597b345dae72570d8610ab824",800]
+["6cf768b531b805fc9ce867e1f420af7e988ebcda56665d3cae1b68304ad3af0e",802]
+["aed08b75a3ed7fa6d88bb67e3fc98c62794b29f4e0ef9852a3b8842a3e434189",832]
+["5bed23327ec02378fce100b48b6507f7ad3e24ac192805ceebcdda221c401969",766]
+["664ac125698cb7e5b02693054890364c2cfbb7f4390c29fc1385815012a5f871",736]
+["72964562f9434dffe673b15bc24afa90fb8b95f6ff5e5e91dadd7536301df16e",798]
+["275f188f6851a8136a3d68cbe6c4f1d06c75c6152d3781a1f34fca9122e344d7",800]
+["f4103b2523731a0061efadb820e6d7c3f3ee95c4ecceed450a2ead015ddb0d33",802]
+["9d784a2e45252ac61c102e95a47b455792e56bfa1583d29bcd1a45674d51883f",832]
+["6a264bf993083eb0af0e6221e1f0456706470910d4db5a304cafb241fdbdbde9",798]
+["12e1594d32afc2d316298866f1370befd97b505c62aebce3f07ce8b846ff2205",800]
+["59d3a82427f9b43ddecab8ba1212f6b2316674a9054944ac1f903d009b79801f",802]
+["ae33dbc21629ccf8a58b0aed61fb74d29eb1ea96e889f1d0bb977171c2b33f01",832]
+["d18ddccc87f52425ab0e1023ccb2a79749b328e1d5a58a13497d01e4fc903038",780]
+["eab3d2b22f306a72b9a26bdc41d28f405884b9e9753340894c6d66913a6b36ed",758]
+["3c6fd8227d6397ef97d00cf176e086aa5a35b92b2d0382b1f63f61cf39c5f3ea",816]
+["9488bd37490f1c19445b3ee7272ce021332402e1cb056e78998d43fe590663fc",854]
+["394dcb10c21c220614d0264a48ea73a902803185967298f061610df23f254475",820]
+["79ed183aa6f1c0d7f9c8c86225c8e512a973100873eef567428e1093e2f34e4f",886]
+["dcadd4d7dc57abad4d46cac4378020ab4e82c6edd6916899d017de119cd29dee",816]
+["7dfc38897fa5307188a19b01807002d0f7d57e80d94d7bd7ac2d3e9551cfcb60",854]
+["0b25f24296bf8300ac6292ea0fc82df1632ef0d34d7e49b653f33bb799b9cd46",820]
+["e8e9b02111e08c255c6b89dcd2218ef43d255a314f3d8051d47cf4e20d7d0fc0",886]
+["b3bd24be8d297f73eafa7bafa07782ffc6fff39fa4ab0494dcd91a9310cc4b71",784]
+["cf77ec7ad10ac5221e2438e7fc52a00436c5df98e461ac3f34aa8427e53f415b",790]
+["7312af8fc79cae039c31d0e3102df7cb2805ed59374dba6e19169185a07cbc3d",756]
+["28659d5d2f3af70d27b3f170b2ca3bedf8b46a25ed55d86c8e3f8c981ac98f85",686]
+["41449105e34d5190599d3d9b367376e4b5e889d9381a71c2de66016fd2a42cfc",792]
+["2aab9a6b95f9bbce8e71d29f7a62cb53dbcf880fb05dff202b08193e6706f19a",782]
+["0a74b8bb3c5b197c206b1125c895acc8a314d9563c25a7fbb64a9a5706e5dfe3",796]
+["6870470ccf81337f77bdaa3145b34dbd359bf6cb47777d7f78c20922c59a578c",814]
+["dff17f19a6c8b251c13bd20d9d9aab6013b3e3fa2e24c5d27160aec654d93d26",792]
+["39945101274702570ba0c931daf7eada823e954b95700987814230ffbdbf4637",782]
+["1831f291a119d97d240459dbc9b16ad80731b26e96133e55bc2aef05838879f0",796]
+["38e1c9432c34d9b4d3b60556c5a986ed1d4eac87aa0588d63bcf9518768c62cb",814]
+["86f7d4273371df3e29e0e4b3a44ab3165f95ac9b56ebafea52e10d4b8c86fe32",760]
+["1e015bf7fec2d7f0fddf33a05302e432b69b077104ab3a96da4a0c7509c8831c",718]
+["3ab50d2a60c8235cf88fcb9edee1d64fe896ec72f094785619a47cdf7c641a25",799]
+["bd55ecff83bc749177c4f2b865794204b8e284c1e5dddaec1234b56f5799ed26",813]
+["432714b7cc11c8549f6d1811f02b83d90629aaac4b300855554626053be45d61",835]
+["4652cbea181dcad0cff42658ccaea9bc233b2ffc90ef33e0721569c63f26dcd0",909]
+["f353fd7cccb0548dfee71bf089cb67a11583496caec5fcee8797d3de46266655",839]
+["5ae1548e0a475598780a52fda099ffa0a0d1a36800249e64af491482997bab51",941]
+["35678ad711cad817d4cb095c4a19bcd9b8a4d3dc424ab4a4bea6049163750f42",835]
+["62b14c53eac48ae1a0b9a145f5fe180e8408ea33dd43d78902e35862d0b757ed",909]
+["2d85718638ebcb23d7cd5d360e7ca627e19867e3ffac648af38f17bf17b057c1",839]
+["f108f988a644af3f16375ed5b8302c1f3aa2834550c2473bac19f59cd9df6d69",941]
+["71800496a1878c68ed28ea6c5818c76a77fa48fa87044c14085e695046ba9003",803]
+["69daf2ab00b3f584fa6a38b2d52185a78453ecc0bf39637995c39ddc60942140",845]
+["08bf19b54816fb2ae0ba1689a463f29913d51a6589000eadde7a9bdb8230d3ba",799]
+["7b33b2e59a082d5dcea12e207b89b6b8af4fee4615a66defbdf07032c84ee9d4",813]
+["e650d285de7976760f211bfadad1aa89e0277c61466bb9427209740555d7f609",835]
+["33ff38e4a37fa435e441b54fa93bf606d1ef8c1429d855fb0d557c19ee851f87",909]
+["08c38e65e33eb17d5f6f464ee6cd51e900ce29e6e74cc8fec1be9a187b435628",839]
+["da2b9c1b2220ec4b101e277ffa5b09a11c5e42037d95daec2b1c74a596e3ae63",941]
+["23356cfe46ef6a245869586b99c76d82d0f2f29b45ad757df93f0a36079564a8",835]
+["fdaf0f3e7608adb23d67cd3ecb7281e9eff6eb68328ba1def887b749ddfc110a",909]
+["dbccd8edd55b8a921a71b66de3a455c61ea6cb61c2381c666675f9b8cd3055d9",839]
+["7b453349e2b131f8362c4a2d39a848004fd510891faefee83f68fbefd2259cbe",941]
+["ebe899fda1d9495b3d7c695434d99ea9f2aa3438f127bc61c64474c6030fe342",803]
+["2087b2059dbcab5162c64b7711082a5524adecd76f783b2371a8b0ee69983d73",845]
+["21d24aa30aeb3ed3a6f72cfd580747dca4423ee939c71624516d7625dc79f8d0",766]
+["18bc6b19e0c3141bc31ec68862362b89dc0d04cbf5f1ec8fde32997a9ed12046",736]
+["b80df83746ac2b9242daba5aa4ef14ff7367862827f66fda25807a0501eff942",770]
+["76b5557dde2f0b5a2715f83c54319c01b5c094787a261c50763abb7713032108",768]
+["7499eff4347892d410572581834dd1d892f591f055b0f3b4ce364ffcb598359c",767]
+["4011d0c5df15479a53c94f51766d6fc94cba60d970771aa1f713ecc247d39abb",663]
+["37e5c1528936e75c2e8423e6511a83693f9e82ccfb57a770915d49787b7617a2",771]
+["4093dd7881d3b2625ba7765ffc68d644df870647708c9c7cc1e6a924695e684c",693]
+["db7dd080d107cb68c6161dd5e3d61e081e1140fd6629e880ac1b1566ab6d103e",729]
+["d3e7a2d45572ad9fdf634ac6b72fb46a4b4cd318a23a8298b818bb2009a02abc",491]
+["77b69624352da0aadbe099b9974df6671a08bd69d5231179486870f1e605b4eb",732]
+["cf15209722be50e7ff9d04e682d1f72b155417f4ab0fdf4c20869a140d5c795a",496]
+["628c83fe3a32be4dfd7715d94443c01d0f05bea01a39fec9955894fff45cd71a",1158]
+["490eebca265b358b6020c17e55d28c68e0990e844b9dfa4f5468e6b7d635bb4e",806]
+["bc549581207bfe3df2cdf90c29772ae05da9d4ce4e5411c24caf6a45c1304258",1145]
+["6b2cdb92a9033a46b7fc73d79854813f4a771ec7c903f3893e8fdeb24196776e",778]
+["96ca8e0f4c829963e350324b0e08fa65d1c16a220a676738b4104b0d074870e1",802]
+["70799acdd32665116b1726c0fab9680a1c22fbd852685038c94aee49c51d578f",653]
+["1ce9abd18d294f8748392e43d5023b349f84fb029ad7b274c0e15533274dde06",789]
+["e2a867ecbd0dab2ad97898c1192a72ab7d7c50dcea01c7f5f521a51fb3c5b263",625]
+["888ba02c3cbda498c03b56884c17a8f675e6e8548066881afefd686b74331839",824]
+["31ed186de642e416ba35a5ed407d984d69cedaae738e0a1dda1b36623acb5a6b",661]
+["e6530ede2e6458a2a5f8f314fee08e115be3584f54ebe2f8bec880587b2ae7c2",811]
+["2ed182ed81b0ee4620fd8b970db10c45fd7133b4f96184444c1c489ec2154eee",633]
+["7c8ce530acdcec84ee6a3db31c8347bf8e3e7704d728c7c73055291f6262742b",1542]
+["1dac6af9f2c90fa660dc1a5ef8593442c65d090e1b60a75299b7dd13c9c53e20",966]
+["0aa352617b9c5a41e948ae42fdd5947c5c72cb895c682fbce95bea3150ae584f",1529]
+["74513f3a37b08bc1b01f46dd9d521d121ac5b729da812d7ee864854cff3aab93",938]
+["2e0571204df49214ef01454b26842b043d82588a0025a4849b35722b89c4c154",868]
+["a423a59bf434c630c82f9985ba988396d5df633294a76b3fff8a2e61e0304472",678]
+["6ff18c129fac48daa9bf54dd73bcf54c44889844f148a7671383d9f832b2da1e",855]
+["a4860f53ef07daa98fca03db2066c41caaabde11887a9d92c9e9c19bc6a45b0d",650]
+["20c2d65641bda6c51c20b8d8368011f42f9970671f073313d58b7f5f2af0bd53",2339]
+["0523d965fbbf796736222aee154a3a09388c1cd3a2d3e5b82a9b5db7c5ff1b37",1318]
+["d5b05df5aa77fe99f31192f26083d81b6dbee19c0661ca926b126df19ff36dd9",2326]
+["f9c9f1bade56eeeffe10b18df2878db7430ba4fdb471a17fe4a9df8ac6af3252",1290]
+["986bd02aef67f216b5b372e2b0ed29234565a3032dab6e892c15d774a1ec27d5",963]
+["0ceeff19a48d5f12fd2712e650204ac24e6b9fae8542479b3c8f477d1882042b",719]
+["61d4209d92b523beaaca3dcb3ca3824e39345e646af0cd8adbd792b886c62c52",950]
+["aa4efb092b8a94106cf53c4b3422debce3894067364e153360e0a5fe67a47adb",691]
+["461ddcfb88d7fda9b17a3f686de93df1c47f73e65a53048d5d13b0ecaed74371",807]
+["5964671b11c7933e9150878e00a37c80ded4ff0ac47b4f48047ff71a9d9dc4a6",621]
+["2cb9372c86abd383cdd69ae55ed4d195b754a058d32ac7a0964957272646d27f",1542]
+["0027f84726fcf9db764fd66103eb2b5643cfe3b02eaf855b93e7dccf3af83630",966]
+["524e6020b1136bf3e43ebe22a161c23f7817a9a5e89703240492044a321177d5",1529]
+["2b772b3efcf3f74c56fc261899e6b4a43e675787b1acf8032f049392c109246d",938]
+["09deda0ec07d00393126f862fc749683f1d64f51f2bdb8a8663a3f17fe87e02f",824]
+["5b5cd921e2df4a3dfeee6c817b189452689b21b1e90416818426570f34ce1b69",661]
+["30fedfab4821977bb27e68f611edba21d6f72b80d09b535887feeacf8c7ab47e",811]
+["95163eaf4208b04bc634bcb1a7e73d4c2de1bd67cd28689637f7f617556cf8d9",633]
+["0945488ae618c5e9d443579567feccfbe8d4f29d8237eb3c29592850229efa0f",868]
+["0b88b050bab6bd62f7b194dc3a446640e7fda0ae2d8e663ee901a155a8e0891a",678]
+["b5377fe94929a837eceaa3d0ff9943e4e32b815f91a6802148e28493a223a737",855]
+["506e2f8be64c29f97739e849937ae803f3c36bfeecf4a6965276e04fe9db4ab8",650]
+["145237d02fc28334fa55c8daab60d7cf4fc59d8dd9e1266b227de5df7a8ed719",2339]
+["928f46626987c0a48b4326448bbb3b6d7079497e90c5c81eca4a1c433db9d66e",1318]
+["dcb413540cdc794d2d2122c828078b037b218c839253d0da7c02732eca9dca36",2326]
+["fe9c8400dfc05dc758be19e1108b47807742f8d3c4955a2de1c03d2c91378513",1290]
+["24a4cd653314bf5c9680e122e2ca6be4e855309194c6b90e72eb29eb3bab8eca",963]
+["cb52a2a974a83d0788a3c2daa70b2bb29fcf43f72c69ae478a09df6685259009",719]
+["941ec36e769bb05b0921d2dc9088b543e45819f893bdc254e09f401d79dcd785",950]
+["760ed9ff612081cd2a9b7422c8082bbd877c68836079695e720bcb28a4661d1a",691]
+["f100a9057c785d52bb0ab0068ab582dce66727e613da1f35656de822acf7aa7f",1155]
+["86936ef51394a0d60af1c16ec2b62b43319c4a9037ddff6f5a36a553f6e5581f",799]
+["31eace610f9e58b733d60fb0207870b8b4d439bbae4e48a2f0b8295eefef5f29",1197]
+["087b71f152fea917799d91d7daf6b1144824a427cb4ce4151f252e45ef2e24e2",777]
+["6b5fc5a9bf21e37c88a0b318b48d38d3fb1c8e68003131a2c19df23e8bab31a7",1142]
+["f0fa7a315fc658d9e1d9d24880c7bc200dc23d99a07fa5647933eaa479980c08",771]
+["640461eedb5f79c26cdbde26d26adb7da89c8a9581360608a407f802b8400ee0",2361]
+["054fd027a4691c20ce879af067136c98f820c4802121ee1c3cfc4899b04ce53e",1336]
+["e1191c5b215ade81103621be38f8c03c328cfa284fa4d4631fcf9c41bb6ac4f8",2348]
+["be0ff6d9e13e3f077911e5d66bd91280f97b7b7e42ac5b95597fe2f7e1983c4d",1308]
+["6bbe5a6c1b8229fd3c94bc655db39ed14659926cec6c552b219427ab0da06abb",984]
+["4900bea3da17029aa1444c402e3955d8fae0f6fa19f65035ad4c0b98c1442c99",737]
+["36cfc8902d47dd1ea769df5d2232db1303ab42eb94256c03f8e214c54d9eb8df",971]
+["4789fefbe5c5df596da0693369ef710265d25b34f1c3a1b3c59833ba17bb42ed",709]
+["85dd1c6b65f3fc422e9a887e28657ac782e38086e806ddffa0351ed9df3e05d9",804]
+["c9c3cec00641956fea75bda2ca3ff57f10a2ec25a93e707d99d4e4e46501d30c",660]
+["e69c53232543c1cd8768469605b5537989bfc8346a4504a19a8c5bfadad3263c",791]
+["ab21cc9aa3bab27ad14f5b7f06707256e711abb17df5de8509af6e61e63d4e96",632]
+["ec1c91af4c4ec03984a91a45d8c6bf44cc300c14d5414174ab7f0b72cf46a0fb",819]
+["cd36fee2eb68af9e394d902b2e89c33a9674dffa7f9b3ed1ac6fc014d77d5274",664]
+["89a1cc7d1732e101835abb4069e59b3d6aa9dd2e38ce9a9b008a761d58858e2d",806]
+["e25fd007498bbc159431687f7a98cd9fe295fd5ac9cc9f554c5781f27b3ada86",636]
+["4ec16c5778e23aac32bc1771a02f0caec86fc0582a8ee97ee2ca2b9aca0babbd",1176]
+["29e1bf12c9b09615eba46f0baf10b0f19d49ce518419f24c9d4cb9b93f057fc0",817]
+["b7be893a908a40a000c51c44c42606afdf6582a433e11f2fe24c6a2d7661f5c5",1163]
+["f990b559f849bdec07f732f65f57e55388538c8c70b0746c9783c84ac79487dd",789]
+["4dc5eafa0d9737e4fb71d5d23a20576c341c35c763329f1ddd817b39c4d3c5e5",841]
+["a67bb15cf4a45a4b5385773273116b320664d7f01707e4d70bebd5180334ec2f",672]
+["464b85313fddd16370cde134e0e2fe739ffb92deb0484f6653f58461c1b6b2bb",828]
+["9b1420fea98defb524b8d2563cd2001bf58656510c43eb402904b93f86b9c4d0",644]
+["67fc15f6aed8080d432076ecdaed4060756eade5c575ead6e86827f700ff4474",1560]
+["6a620d7f2240a4710dece5f0c7c3b7c29218cc07a949f6f10ce2649ca2d0a537",977]
+["23c922883f477a05cd69a540f6b5a11ec30d2c735a9c3dfa8e10c0d71d244760",1547]
+["cb5500a97a7f2a20115c2e9b54307ab0cc2e5affd5b24f752a683368dfd81a7b",949]
+["ba8cdf4f76cdd57d0ee8c9f80faf273a896a8bdfba37b52b94c624b04861b920",885]
+["634511a1f5ebbf4a231960bc555eb961f30c636a099feffca125f7ac33b32e5a",689]
+["f55be8879bda61a2ba08cb8bbedeffb1c49feb978f9116f857ad07c5f234d4de",872]
+["d82d4e845f0a83d2a41b9e8e2ed24bc0f112ba1f4aff5691b3e04bef1282f1aa",661]
+["8c0033d21d02e0f88f706e927d129ac74dd66ebde606fd37c9126b1bceb531e5",2339]
+["f3eb74eb84d25674469fb0dcb559edb279b275404cfcda80fcdffdc7b0f059f3",1318]
+["3d1c097069d74a87daf094ad389ee8d6e3a9ec6617466d6931f24bf0c6cdfd8c",2326]
+["b774dfecfd9ef6fdc1ed52d5971511905f172be5ca5dbf2511aa886717dd716d",1290]
+["571b8827da4b11ceb1ed023c2921bb2614b6b2e75ad1bb4d26389963809cc732",963]
+["70aba561fca3194bdf81aca7d3dd70d2346c1c27be17e1ea764361bb148e64d0",719]
+["e2a0ec72f13e818fada2a1fcf592d1e0044e9fd25bc456748f21e0b8185432c5",950]
+["3db0c79c39bf3db9cbca1d3de4c0f8c91af165d40bedbf0230c4fde0f858396e",691]
+["7842db20f5695cad20a07e3df833a5bfd2e5afd1265ee59b61d11ed833194dce",784]
+["10f347abb9fce161010b7b19582e2783540536339cb6e5bb88efa756ff2edd07",642]
+["4f90768f3d461ef678c46095273d1fa0e1541b5f8d9ec14b49d24e21d732f674",771]
+["5b9183933c8bdeec0b3fc66750b4bac94928feb9b7b29262ede841b56f746f77",614]
+["7e8f1357ea46acbbfc14797fc06d3cfc3b17830137c639382624cdf421c6d8a0",799]
+["9ee15ede6cd215ae0ea95fd05f0426c1bd6dcf6a7dd5f660f61e29147f5e8876",646]
+["579d1e8fa2f5cabd51d09dd2a29fc43082e64a8464fb532d3fc47b612668726e",786]
+["311ceb4c42548e9a07006a7a2c0e56ead2ab470ee0795d9ba81104d84d15f769",618]
+["638b9f016757e8c303a3ed8ed07fdcbbc37a0ce9afd53e2a474f7dec7e6b93f8",1155]
+["2a25f150f4bc3fdac3c65c907c28d9c6147f6ab2d6f0337b59ecc4a9681b8966",799]
+["bbbca340ba9640a61cbfa771e9c5a4963a1e147582f408216ba2519203c7b8bb",1142]
+["abaab38c1ae7861a9927420d0b1ea6eaa89cc162253ae0282e352700a49a1091",771]
+["e4be4eb52593e7135fa4e11d5851d0b15639694c51d7faa004ca532eb95365e2",821]
+["64a874a0412f541689067d71bdef856c048e504762dddc63e8f8301fe04987fa",654]
+["1b52dd3b47386026f874c662351f20938f5e77974e3597eb5d293d6940bdcaf5",808]
+["b94bae0fc28a9db035624a5d3ff4dd4c1362e4e1ebf3d3d43fb553a7270566a1",626]
+["efb8e23cc86f37a65318624d87fe7876c3a6ba0e6d70688d90f01030384f1034",1539]
+["cc6e86a1faca5f20e84412b8e221ca6255975d4e7cde5811f0c7ce560cd625a9",959]
+["e96904d78fb3a712a3930b816acc04a25bce37ec19becd3a45637ca68c179200",1526]
+["f8dd556d2e00d73d4431151b1f3404e2ea0e9c5a56e65433a58a2c39a0032d80",931]
+["3402bc1dc7c70ece1c92f349e6f0f50daa77bad97ef6dfa1659ce80d74d8fb75",865]
+["003db43102df4873d9648a5ce086671548d18147d160346a74b3fb42a78528dc",671]
+["dd59ee80be9fcc5fc1a58b7923ebcb5848921b524a62a5a031c15cd9d5493b03",852]
+["1bc9a86b4ea4466541b98560f5c80cf5b4a2ecd7cba15ba2dcecbc1a438057d2",643]
+["ac3f54566c4ddc7ece5c8d4fc4623ca06071d474ce5f502f0eafd0426c9c099d",2339]
+["ba7e20dda8bfcc85a06c9cb484d619fdc524543670f33966c142950bebf8df28",1318]
+["002f623ca8bb8106f2c2994d4a2355629ef27b3e58d9cc332b79bb5fd2f8098c",2326]
+["c5133ecf9d1a3df46fbe320e15904889d0850bb0baf2e5a645b10dee1f6729b1",1290]
+["117f193c34e2bdcecb414772d53b094601c8f0aa2ecb5c86ebbeaa0a9022979f",963]
+["eeda0b026e43b3aea12b10c0bbc9638914f09aae6827813e52ae183492acef66",719]
+["bca2dd8afa8b8cd4bf99f1475a042430dff8eaaa1a889ca532d156ae1ba3f0bb",950]
+["e6f2718acf4e44221a2181d3cac0d72f0547da3a5434ab63bc0f045277ca3a99",691]
+["f4115b74ef4ff79d7645bd859393b138d45770f1882512ca51221d0cfcb93228",784]
+["51147762d59fe76e6aa63ff66c4180c7c292e773dfc292d2a7649444c69bf689",642]
+["f22e644ade2783dfe8e4be617762b85ea4fc3156f922d9c236974efa4262d9ce",771]
+["c3b9de68730ea37a95a2add0833079a6495217aeb52030afffb30018a4288879",614]
+["68c909fbc3f59ea7ca13aa26ef408da56fe8797a3f9ef055f259f0257827a023",799]
+["2fa31a8819d12f5d6fa330039572c2595ca513fb8fb3cad46ef1ffdf02547dac",646]
+["ae0c5363a7a47541a54e3546a11bb5ee2848ba2e6a790a30cbde6e7a1e185c2b",786]
+["a57a194a68b827946dc57427c4fa9495d82ef299e50bf5b4e99c6736ada486bd",618]
+["393f0de2b477cb06ddd923941f950ed1ea825620f4e22419a0b831a1228482e8",1155]
+["6ac108e725e9162d254ff96a3db80cd54ef4a9422f4ae04abcf846e65e8c6fa9",799]
+["97595859bdfa9e9b23938387e4930e557ec3debfce3cbb5e8d4955a2328287cf",1142]
+["af4e408db0de0f315d470047452b9c91eb1b79df8f9f1a7f468702be9f8c582e",771]
+["9ba63411586021de5f076e07712b4c683897fd77af7898e5e9780b3f243aaecf",821]
+["af3bccda5170276357829b1bdc0a540b27e193fd82367c4d31b6a3820082473f",654]
+["2826be06190eca3a34195333f53c7c4431c6d3159990b033f731f438df7ef4a7",808]
+["a7941b85ae44dc57233a3b0cee5e4e2db55230cabd09981466ac950394d71dd2",626]
+["6548514093fa25d83a6de5add4d38609a31717a79724f0a0dfdcb33ab009bdde",1539]
+["11f9f40c7a42144ea562d3542f052d4517cc6973daeb9144dcf043b7dac33911",959]
+["d9b43566ed8ad3a1f5db49c0da268a66ee4d823df74a029272d864f35380bc75",1526]
+["0f701321d945c8892a035e1856eafdc32871d8a7aefe145bd35939a401ee1eae",931]
+["78a307cb7ae249bd0c73abe2ea3a0289331b028736d2fc97d1986d946bf43ef8",865]
+["0f026057bccf5781cb608f08029e07b9886d45677f70f30ecde021fead14a118",671]
+["ba45cf5bb55b3ba5d340711ef000756102371679f519586777a460d79e6970cc",852]
+["cc7c966235ae9cbdfe0fcc118ab04bd55db88f4ad8fbde3e2cfd75edda715945",643]
+["0843511df49371725a5f9ee77ec02bdd941c8e41a81aac6035089ee221d9c412",696]
+["49c2077309d9ecf34baf2abb6c868e16a661eaf322fcc78e03c890cc126b4c16",432]
+["4a5ffd0c3dfa424e7ecf5f29915533693e4c3e8bc8caf14a3c5113ad72235bb1",962]
+["570c4ad349de196b0f179069060b653682833ce8b3c81fe5f728ea0a7720024b",1093]
+["b5558fa2c6d1d59c2216ac613b505a911f273251a15070cb162f1ff47a83d825",962]
+["9ed25cedf75d054de9d9fe6f4d29948a11c1ae7408d3fe258989fc7aad57df65",1093]
+["dad7a6f462b8318faf8bd2d406407f082db386680c8f68b3b543dc19ef5963e1",962]
+["4f6dd8325d9463556acce1425a945aa5cfc6c8ce72205ee333fbb05ae795678c",1093]
+["ca6f70027e4d458098e73103d9fcf4d22b7237a43835d67d516165e7a42016f5",962]
+["cdd2c172e52e20ca7e61575a6494f80bd312bddccb7df6b8ea5776cb84e251f2",1093]
+["9bb2a785ebb87092cab342e6057452f8f0f71c49074f84eada8b86861da86da8",962]
+["a520388f7fa7895d8da2fdb7509fdc8775317409268b8f36dbe339abb77772d1",1093]
+["f64f35f625f29249e6aa566b687948d9de772ddcd6665d76770507b1437ce6aa",962]
+["8d9d36433283ffba769f9b56365e8daf4d411c95baeabca49fba8306fafd39b2",1093]
+["0accc0eb67d3496f047cb821d659bf041eb722220a44997360b7db436ed36b37",962]
+["5973076599a3efb2014124b6f70fbad74b815cf9b5566e9df6f01a84633b8e0b",1093]
+["88e560faf2fe409e600e973d2e96f0d214cd19793c2691a605a3cc3ff02aec82",962]
+["3660153591a029df2cf540bf1c9349ebe3287df64f425432cdaef536475b0ac1",1093]
+["3b3efc4b7b03f1e876c3efc2ba6cf1fdb838ebc0835c813bacdab543ceeec9c5",962]
+["bd71dea22ce28edfe7c7cb138a445b4175d80ddd61fd06e9e55663f92118c356",1093]
+["a9b6c4515f1e48b119400adb013b48d9d055d6c98fb073564638f8c4feebb895",962]
+["dc6e22f6cbbe071a65edcc12dd6eb30a9057cc000f0362bf222e0ce6e61553a3",1093]
+["50b232d4602ad2117e9c215dd6ebdaa22dcea0a027190a01e68795af3e8396f3",962]
+["ddfa171d75f87c9fd6a4f9992c7a993490fb85a294affc6140f53bdd8bab9518",1093]
+["de50cfda82511e95b4817f9d25484cf54dcad8b349abb12f7d1cd89bfde549e3",962]
+["7bdbd73b8648cc4bacccb72c5b3d55eb486b3a0f0a207ff245a2711a1745d4e6",1093]
+["1c7c32240f1203a97ceb81d20c0fa7680f2be3622604da6e3c862127fc0bd996",962]
+["b79d6194dba913edc25d403d4af756bdd4627ee1c7bfc81869dda270f819040c",1093]
+["49e05cd545ae0f1fa61f287bbae2788ff0b650e1bcd803cba4d57c1d69e44c75",962]
+["678b75ab50d883fe2eff72d56d5841483ccec36dddc3c0cdea01d26cf759b4ef",1093]
+["ec44f27db8a384ff227486a4a72158db8dbbd5bae34e71568c959fc91e002ac7",962]
+["cdf806c3438f9ffaa06a64a313191c49304f27244ab59961348bffb233b41834",1093]
+["1bb00f463812c4861c98980854214143c6774a1e89eaad3010b7236d213766e5",962]
+["8631afaa3e6ce554eafdf0c10fecb0b7b5d2c84dd7751c868afb82c4fc698b87",1093]
+["21d3d1a2314b7d896ebbcaa54fa1947203eee70eb22195c5ee8ddb6065eb442d",962]
+["e5bc19d01eee78b8ee52a6c1cae01186e3348dd44c7b1570cb3de91636f255bf",1093]
+["998ff1907b40275598bbaac2658585dcbaa53baabcff3725419c12d488d3601c",962]
+["fb8c6cd4f67bfbacf33ea5ffabe3f6007ee1e4355effa4e0e57b85b57db97a32",1093]
+["0030d5bbc2a3803407d4cc08dc31731b2bc2860a7b1e26cfb963ad8c4791d88a",962]
+["94c3d001e05f2b470ca9a8f29442456d32f25cbffa18bfe36b03fb0c8fd69f49",1093]
+["d76d3a4f47e1703f9e4d640b37c486588ea04e648b82ac9d797d978864ace426",962]
+["cbec4baeb73aeae51248a9e4d6b7470937832e8592d432a9a47266704e637c12",1093]
+["e51ed807823ebd7aa1dc25177e8073f2a5d2a3b855d2f24e4e4a908cca9b57f5",962]
+["e0c2dc081529108e5637fbe0088f523db8cf3e76c4aaec6d8d97330d85a5edda",1093]
+["5507a93005d60157e0218f3a1f45683c97bd9f0b3ee0a5f729440f9aa91aa58f",962]
+["3972aff5fd6476bfaf5cea76f6eba5244a0d190d0cf54cc9f314773834b9f1ec",1093]
+["07b1d03ea6d4febf19eae2682fc698ec20b88d0fc5abe4f92ed15f56cd039f4b",962]
+["0feee2e571b0014ec82e5693b19b855b62c8113d59c979ae1eb023aa9e90b21c",1093]
+["562517367b3da05f37e6fb7ea7f824fbdfdddc7ef2c2a3a7f3e8199ca57f5d7b",962]
+["5b387f41a34a9cefdb93f0280468ad9200e0904913324537f4c44e1b6020f801",1093]
+["0bbbc8f1d95ce17fe444fa215838e843fc97c41b4ccf712ec0499517a855759f",962]
+["61fb0db3130fb73a937d471e06dc474605f64127790198958e8c3b7a62f33a39",1093]
+["aad32f9ce417bbf7a318b7e6a02197abab0f0d122efa6835c205408bfd926d8b",962]
+["7e377e18c8a6df6e3018e1b68ff4583ad14d28ebd21212d6281538ea3f55eada",1093]
+["0c8cf81c74c7e240823d66210dcacbecd878210ce74432614dfe117f8364793b",962]
+["f863aa1c17b98e2a21d5ba2df7c23c83e25890b93a8898a350d5aed991d07dce",1093]
+["3fdd89a48579bb7cf416e4a204648b66a1e222981b8f9de7d2492dee5a314048",962]
+["3bb6c8761b5edf92c41cfcee5731698b71918073311bdb67234f301890147557",1093]
+["82dfb1d072b439ea9aad9040c254ec922156dcb49144bc7419c6f9a607aafed8",962]
+["d740acd49c2873be024b262d8c4ed80f453ccbabdedc410621b2c8607f754075",1093]
+["9343d8b4dcfc2aeb6851f23abf54653d4d503ed45ea54aebb23788f12fb3fa0f",962]
+["69d997cf918702f2d97d26176c472741eaabf30dfc1d230053a8f31eb3e071c5",1093]
+["5aad3a10c2d1cdea6f796df5c2c08c316a794f5306a06188a5a2056cc50223ab",962]
+["5a415b93c47238cf63d67826ecfcf3f1f90ca21981260ecd957c6bc9d775ba13",1093]
+["feedf5b86fbf6d504c9043765445ffdebc769c6149b4b0a3d00b7e9fdd70c8f9",962]
+["d0d18bc1aaea706d4c62eb862140c4ddae45bdfe2ff5230fc77688b9343f8ece",1093]
+["4870fc92bf0a5439d34e76b13c407f86123221c552bfe0aa3743504d9cf13cd2",962]
+["6f0b57b0d6bb2ed539b7e29bfb5f44439923d7e319c92924982b0f0f855f6ab8",1093]
+["1cbe2f6ca782cfccb81edfd0cad570fcd3e2522577cb3fcdd212e4e43a46832e",962]
+["9f1feb11365974d53fc98da8c40d8e1081649e6b0c5b5bef672f64282d91d2c2",1093]
+["2ef3a666f3575ba8588ab732accba74ed5b992941e1e73e784e6e69b59ec6b40",962]
+["0e8b0070d5cd494f3b6f5d2971b276d7abad8cbff8b255340aedc9bb31932fbd",1093]
+["96a8ae0e71038f71b1556639b7927615e52434064476e6e72b70dab66da954c2",962]
+["1c4c4826ac32d90e766ed0a2b2ac5e98e3be129101f9ca2c36c119d759a97884",1093]
+["73d363dcb00149d0ac6a518e4c296bb5ece9350b222666577579f025e56c2053",962]
+["78b2d0c6025bcc0eea50d505aa81be0e34d105e9a29a5ad0ee5aa82350fa2357",1093]
+["14586f6143b80aabc10bae5981a373173d6a0ebc5c1f409141b946c6991559f3",962]
+["8677d330296d1d86855deb6db645f1a2bbe582a20493bc72a2823fdb5da1df2b",1093]
+["9560e1bf9b70f60266db4a0b5ab130c16cbca5128c5fb58b0103a7adbe71d453",962]
+["cf10fcdd545d8cc5c69498d2e3d6bef26d591dcbe25069d9c342d9662f4ba391",1093]
+["057e18c1397043d65c6413877bf8973d4cd418d648ab46f59f7642928555882a",962]
+["8ba77a3094a0e4a8b2269c5cf99ebd880772d21303200e572f07197f9a468be8",1093]
+["ed1936a85244b1b429a29b8bad280de0a0e89a73ff1ab1bea10a0c9989835b95",980]
+["cc6a9838cf7b77ff944a884c81e07ce894ff0f288c9f352a4a11705806604b86",1102]
+["d7f180129c9229bcb07c43987761e9afdd96f84dd323d60c4a6db3086824e6ce",1000]
+["37d5be6a53a5391f3a1d6aad8ca24fe81eb18d9b7b010748fab2c42e4ebca2de",1172]
+["bb4b9eb53d3dbdf55a7a134c47fec4fc84c78e7a5b11f06e8b9fa52e59ca2aed",961]
+["0863b74543ce6d274c04f84989cb1a4cc7de836040f2fefea61e03f4c69eda77",1117]
+["06d1975b4f6774cdd0723aecf76f58eec83e63ad8b140c3b8a38ecc3572bc4db",961]
+["1da23867f55047f6f3fc20e4305e6504e6b6a8d308349d341a3c3c8202a33bac",1117]
+["a0802322c00baca0b87a0d84482c36d68cc6665cde75cf7adf6e68f22967ab85",961]
+["0e6b2d7c5c08f9e8b9897ef7e3ba4b52fde84f016876810fcade0609acd58cce",1117]
+["6b03f37e06f8f6856ea7588a6ad6bb0a46f115b081e74d44787fd692802eb04f",961]
+["0f6f311269ee3570ed91e7708a90b1f46dbead8d27e8b481f9a594b4d548f5f8",1117]
+["0ac09f24674c9f205a5d61e15d71b68204eca061fc362f79bb146ce1aec83d19",961]
+["4dd7c0fb65c6d58c5f00f33f3885bba0f7c39300f8860b4fa9b17becc925aad7",1117]
+["4199b83c1de80d78cc6d3d5e6edd7c3a2f06c4813880e0f7929fb9e3f993778a",961]
+["5eb5fe4b9f40886074978842a341b477132825f9246780e306381936db65ed25",1117]
+["365634754b7d63fbae155292962918973aa68781a9b4ea8426a0ed840fd47049",961]
+["e806ae24d4e60699d8e30010c16136b24719f9f07242e5d1f0b0e7ca3e873541",1117]
+["d8aa4fe4596208ee3f499fb25fb1a339a29a530116ba3eab01077332bab96a27",961]
+["9e247e29ea0af0dca5bacd6cf34aa34df97dd15832366151e08e63454be46d4f",1117]
+["ef6dcc0f7223f47626b95007ea3b361f1e5d2bbe9addc36999a34806f0935f73",1018]
+["58b31cbc66f525eca25cc5196d0d06cf9a06c597d77638590f7023950938d8c9",1278]
+["2fb763ce2e4cce198eda5c4d21724df09936a631ae9d4c11a27256680dfd2093",1018]
+["8d903a84cbc86b8fb9a69d9680fed64d53e48b078385f3b567dfa24477aa108e",1278]
+["b897e7776b02d0a674b115098dcbf0991eba07578229adbfd31c60f4ee7d9b2f",1018]
+["2028b65ca3061ae93fd0e122c39d14878ed405daeeda47611b26a44a83686390",1278]
+["ee511c016df36d16db78d59032b249f4dfbdae70ed1a259092c5bb08b450420f",1018]
+["594e5eb43ffe84e8ddfdceff9ff38ecba07f2613765b067906e8be720dc23a76",1278]
+["8ea689850fe3cb7025710bde66f370ffa20dbc3354961ace9ee16e4a1a2e031c",1018]
+["104e956220d4ed813d2c5a43baea86e020a095b5d7f51e0f0ed8b8f6b06160b2",1278]
+["089a9ac232b828cde1668ed511e1baa2fca544094161fe1678c895bdcc6e9c50",1018]
+["602568f83c419d32b17e37e9df286f3ca8c290057ceb740a98147b7f1f627ffa",1278]
+["71a0c54886a90ce16ff1f2a6bd3e1f5d488697249ddb810bd3d0b73f5988c8d3",1018]
+["5b578ae53db4229cdc1450cc498b60e4567e8e2d9f3e157524dc38fb1ed10750",1278]
+["d15af1aebb8dee371952978411d117279810ffb004739df3d6750a099c8729e3",1018]
+["a34d3276f652d97b2a30562adc0b40faf3460b4a0e7f281badac32a24033cb9d",1278]
+["b74cdc16adc33a92f2c9e2af16b392ee0a319967baeb443c7c09f8812dc12974",1020]
+["2ebd294d264aef53738f796f0607389d3cf7ba0a5861f3d157a7f4550acfd44a",1278]
+["e0e218d8da1131293ec16d4fe760096fae8c7f3be6fccd023d4c849253ebe64e",1020]
+["f77aed4b033997e0623b9e101cbff1cf6c34e56c8f178484cc75b5f06283a142",1278]
+["f8885637b6e62827e33b769dc92ce9080acf6cc14f6f855b5f636f321649a310",1020]
+["27c3a8a6072c84c54003f917f467ba4d0ca5986823ae56f4496860794a1ea8ea",1278]
+["a17de1640272e0aa653bc1e688dd995a1ba73454306725cbfa2f13adca273db2",1020]
+["b1330dd6768421b51e5e17c63931b8e20bdff9f03c17aa24eef4670484274480",1278]
+["948733611c727d8f1d709532be5f4ff223a62791d401d225b670b13401d36d05",1020]
+["511aaaba3a8f4de71fa6a5e819beaa0b3b2a0ade30845ab8afa709c2d209e56b",1278]
+["20ebf5f5fdb45fe09905c2ea25b8758977602d067fa2f9a65fb4cbec9dabd247",1020]
+["37fa2ffb14f488ee75657090cda1e2a89bc2268e745e7d21f9d6d4842a7e52b9",1278]
+["355ba16323b8c9a514c636d0242271f87e8dd3d6ac1fc3304bd97ba3033100bf",1020]
+["5281cb135a339b10564c12935eeaa6ab4387689a0496866b62c509ecf5c8c180",1278]
+["96b7e14e1a3f33a3f962e0108d4517825142042e18b39a54ecfc4dc57f6fe090",1020]
+["5e814b8d9c6cec30f9ee5deec6014111c2da651de837e3b6be637ad47809ae15",1278]
+["06589523cacb922a3bfe954439724c6282cfff8dd67dbc4f179b6f04e93f8842",1041]
+["7076c985c57b915aba6e815be141d436dc9625d08b5ae5b82e83642f85d834a5",1397]
+["be89581028599c724a9f6448f591c0c67f7cbe0138420b76b5445d90290d17de",1041]
+["64d993d08db84c544d8f10286622b9211ea02a255437aa8fb3126ef2670b776a",1397]
+["4b8b28451d57d313af299992a64e497c5922f9d815dc404d0d623dd6d652624c",1041]
+["e7381d57fe16098adca5176711e8c5aace1db71350552878f2ce3e50882ce30c",1397]
+["27394ff25f264bd689c9505bcdc182787ed07b50b589b80c05d16576bea5950b",1041]
+["5fce8ddc44eec630094e427010e5fa607763454b9a2ca1b79e8e7eaba72b4293",1397]
+["13830d307755978be7311e9be26f0570b89c3e087156929bb0a2ad9ee0c71a7a",1041]
+["3b777955f7ffeecdeb1df7d0cd5a9ab007c8dc40a1fec2f0cbcbcccf8026f803",1397]
+["c9ba73e73689afc61f8891ae645376dc4265360aff9e83c127ee34f7c1f02fe9",1041]
+["fe0525023b67f431b83f565bcc4baaf1ba893eeb04ba50504a2cedb3237ebaa5",1397]
+["84aa4779662f2be8ce738d456a60ab837054cec3f4adb0ea2fbecbe9819ce02b",1041]
+["892b285000d52abeeebd38869450beb0a9bf5fc384b9f19b8c94954b55b15335",1397]
+["e2a59c3560071222d30309ddb7aef1ccc109ebee83b8a0b19270ca529af1775a",1041]
+["8922ead6b608c12c6ef62a3f2e9796a8d0a2a19e14b634eec8b5a4b7c94f17c9",1397]
+["6fe48cbbf42970e8dcba574277c378beac28c373a5d3d814e49b67a256e8417b",963]
+["b4301ad9f65e1a1a372ff984b76618d56c16cdda89676eb59057d37921be7552",1117]
+["3cd25cac46277a8dfdba79ec97413b11aa01dac661ddbc6375d148a318d9fffe",963]
+["27832c8611c1b0e9b0de4f105d9068587a31da9b89992b170e43d67c20213700",1117]
+["268dc441cae050c321c8fe5b25e34c8c0c8514332d4daea33c566a3ad95ab93d",963]
+["2e9048a79b9adac1a8c1d3b4cb60f820a926654eb0d9d996a15e8845c4c0d1a4",1117]
+["fed317cfe124f14281e5df94151697b0c25b6ff0629f90ca10a7902b245402c4",963]
+["d437e5cbd2bcfcc536fc73a4caf1de3afc16303a12e31bb361e269a6eadff2bb",1117]
+["a84fafb36cdd0888125dcbcd33ac2431d0370b9b58f3c24d6ab159adcacfa4fe",963]
+["1334db40729aad305c388e7ee5fc6f51e19cadfecbed5d692cc5be91c5e328ba",1117]
+["d4a3cacdd4656f307f1df8f727d1a12ccd41cac5723dc88b3d7d8134b7036b85",963]
+["f6ad0ec8f461ca610649b546e8d4327dc4d18d524d64f5e70e6199df92a24229",1117]
+["4faada611bd35f4eb80d1e7629fcdfd102964d338d706818074b7ae7f0c03831",963]
+["7eff48347df141b246b97e0d6543f5261218f70f7c4d3b9cb1efb2196536f401",1117]
+["5615d62e1642785d2d8e67bdaef1e6de495cbcc86d4058961788bd0f3b4bc16d",963]
+["ae3a0690415814259acca744ff613f9d80348ab8c6b1170a7f281a7d312663df",1117]
+["24bfac17292c9e46902d722c8fb8797ce025d48791857bc801acf209458c82a6",984]
+["4c1707316861960f8b8ed3c54c6843bdc8f1b57e2daa83043ecd75949eb603e7",1222]
+["0efddf1db45f7b2b831b3e186ce9ff9b1bc5d90316004696bd26ee6926dda260",984]
+["8700da82fd5a83496405de6e311175c928b4fe40412e39ed6b91e420cad0390a",1222]
+["72825e7ef739c48e86da3ae2bcd3c32c514f8dcaff37509380e7845e002ddd03",984]
+["8502f9b8a5287a76271104334cffce437b9564146dc2a4190721840d833a6c1f",1222]
+["c8ecfdb65b167706682668559857f3a34ef2a1a4de61e177ff2eeadc59a74b40",984]
+["5aa9fa691018a02fad4707615b771fac453b4f9b50193c2d63a761335e3020e2",1222]
+["f2af81951a1c0e00aec0202b40f066487a5bc0783680f0dbe69e36ba275281a0",984]
+["6222b1906abea8973b5c9ce6ceefbdb547a4ccc4509f29d93cfba51960df87ed",1222]
+["d55fa257891278852f5d678d3eaaa52607cb132427923000f4337d9f958eee6d",984]
+["edc09052db155a2b1ac6bdc3e10b43daa66295097217780d12363a926a5241d9",1222]
+["b248ca426c7d48862148143ab0eac2f135105a4fd6fee79e50d39a6404483510",984]
+["08f451a6ad0f6878c387776725ad43a645286600d88982241e9ca3bdf0eb3b0b",1222]
+["5896fc143f738447b12e131e3a3e556f784aa9585f327d61323ac0bf02ff1d27",984]
+["e45e91c4aec9cc367e644bbde0930633a092dd5aa9be1135d863dfc50c04b540",1222]
+["8c16dca30d9fb137d3d02f174f5f9e7c568e5028eeaaf61e73ce744b5fb4a195",956]
+["e8d821b0936dbff721481a76f3f2c5ff1a99e43d03b2c0cb8dee39fb9f985d79",1218]
+["e557b08b40279bbbb94adbd11a3ce53fda10cd02b1686c5bc7c8116cd5c4dfd1",994]
+["acb97c8532982b09ca346d50b0a760c7d1b8b85144d745badc6bfe62b387af7a",1186]
+["1d70183b78b864fd55f60f719b472071831092ea3a3d96873ce1296f82f66f77",994]
+["9b61ebb3e531d552ee68520ff100f5ccd7f2ddb1684799ee179b2255a2e1d7c7",1186]
+["c4d542995fd7d4e043de34ba025d5e1cb2475575afd3db46dead59da05f8f3c4",994]
+["6bbc63e72ca2d939acac7671379221088e47d53d7e50e80418c8b671bb25ca61",1186]
+["7e450723c4919fda2d001da74210db72ae3b0772a0ac20b87142e18cffae7def",994]
+["3e82953c0fdabdb81394596cd1da355827af26d2b1556ee106b8955364152268",1186]
+["e0efcd88c2fee0f4008d85e8b610adb28cd38a627e5fc7f9ef5ef059b8ff5e39",994]
+["eedaadea310d9b9eccf2762a18e386247c66a6d4bf1c31cf0be00bd716eaf332",1186]
+["dc757df2e5801ae85a642affe32d17fa0bfd69f95365bba1bf1f0403c7a38984",994]
+["0e2fc789b22103b8bc3a3267cc3384bc532ae1181a1459520a72b775eaf7e9d1",1186]
+["85f9e05cf845660ace7c44d8f16a91e7d812d41b517340ae02058b5d0496d7dc",994]
+["79b2c2d29996d5e3124f6a10f65a41db87a0fc8ffd5e04ad40aed4f741f3e61e",1186]
+["2a7a95aa9f3bde77c565933ae4c98322931e98effec9a0626cade9a58877c4f5",994]
+["95c747671c5482fed977a0fd767caf8cd7d6b0006e4a342cd973566e2b5d9e82",1186]
+["5fec26cb8d6f3877d7535a5bf3ae32f74d2fada0f5a8aabe615e642687ac9db5",994]
+["e7b2eb3d701d21c8fcd49634d58b74f5a3f243ad90fccedf5b08f3ae90552494",1186]
+["67ffbbbcd41915e376f651d2080b5c1dd665ec5b0dd7f4dda6a9e66a76beed1b",994]
+["ac7d20daea443fe5c35a3a9cf15cac957b54a8d7f1940ce3024289c0a66adbcc",1186]
+["9dd09a68aa7668c6b5db691fd1533497dd2095fd7747587a7f5ef57d76f92c3e",994]
+["3d1e4925b320f8ed47e96639778af126dd6c741aad3fc4ca168603eee4caf8a0",1186]
+["12bbb5546293be26e249adfedd91213be47a8b70174668d96e713bf5514dd8b5",994]
+["86030c990547a5b14ecbbb1e8626ded7323e501433b843ef67c0a1b6868058da",1186]
+["0b058bb88a03e57a4cdf5f4473a004d8b5a739e3b58eb0f2ab816d5649ea7d73",994]
+["72adf2e99edd6fd9cefcb21edd171106380be87588230874d9fc790f04d8a073",1186]
+["871354da56ed6406ab67dc599ed3616f8efa8206ab9cf85b82f6cccd1cf37dd1",994]
+["ea62e97647f5b72c73a306a28b4d473106ce116fe1d9ab9801a6d76273cef351",1186]
+["5f5e357156a5cbac9eec42c7c8d757d157ecfb71e0bb4773cd02207580688a3b",994]
+["578682c27ad7f0911d32edb909265ce427a25c63e06283dfc60dccfdd01e12f1",1186]
+["6c37a86e1847475588dbdddef17ab2688ffeb2841004cca15ad837e73a13a037",994]
+["ab5b080948a06ca4f5033f715c2875999fdc8c2cc10bf486d678613fbef390f1",1186]
+["5bc5ad2a50e24ba02b196e365b6390af5800a8c2e1c30c724583e3bb26c77446",994]
+["4a60bdeb9a43ef6009e07c425b509d08757a07b32e1fa6b5f45b032492ba66a1",1186]
+["c5d6a865c5583ba36c47fbeee9327767c3a2847b8ca66db963f2d21cce5131a3",994]
+["7b04393b126f54114bfd0ebfc7a1552253180b5694d45cfd31c1c5078ff23ff5",1186]
+["98c6c16d1874143407f35635bf87f693ea0343c0031bdd9fccfacbb475875ad1",994]
+["af83118966688b92577d868ca59e417796c92dc64a0f3a18cc9cb3e676e802c6",1186]
+["ce9b8961b48e08c834dc89b38c1c26aa40bc00801d6ea46c63d14cda43d57d4b",994]
+["a9a333d0ea702fd0d9c7ced1210d20d78eeced99e53c7ec3d1a607627adaf6cb",1186]
+["d26d4e309fa57f11171cfd0ffac695728b7733d328e75577598e5113914cad8b",994]
+["cfbbf89102c115cb77d0cde8fcc1a0637b96d71bb88be2a2915f35f67d2db8ea",1186]
+["a7e467a3349d3db09d7f0a45cb259f06d148883868703beb57abd3e01721a761",994]
+["19131d0c79736177029f30f8e5ca3c918f0d95dbd14b70c3ea97f8d3d8c96bb3",1186]
+["1d03014564212d86cb725d32556907e0f78793ff681d9bcfcd685dfc56374fe3",994]
+["63469a3168734a8bab9589cd4ec228ffc285d5352a5026a7bbe661722532d235",1186]
+["6552e5e1363486ad1ea1d2164c2bf0ac1448c415e6cbd9849725d1b0afeda7eb",994]
+["2f9c1b1476834b345e7a315354c6d951053e5f98c3df48f6b01b8c13810b5893",1186]
+["c84112f614062c1b1faec63203e7314c07faeca4a423da4257f91b6d816c94c6",994]
+["3babdce72b0de758609dd342a03228027db61c36ac90ff55612144154a6943d0",1186]
+["0383542a01c15151354cad627b94c6dc81b0f0bc34e017d5547cfe8562fdad58",994]
+["e117b15d985ba8381f4d16cb7515dcddd2d793c1dadde2bdc984a654674657c1",1186]
+["71be3955d37446f21417417ff0afa3e50b821cc3a599662a16f1f02ee3b37f27",994]
+["c8bcd3265dfa2d1f6118fa3717d330e97ffaff39ccfebfc28c160c1c4f21eade",1186]
+["ee41441d1b1656949513ba8d25b59a048e9f287a7901a621390aaeb17a96c997",994]
+["381ffb505d8c849b21b936d459e099d333049cdfa22ecfc366a680c519874b5c",1186]
+["26b1622a4dbd5b244ebe0116f05d0cf86095a4014ed45a7262afbc6818a21b44",994]
+["51dfdce732c305dd7e1490b1b98b52f54a8d6432e987d889135093910a8a2d7c",1186]
+["663e20e3c6f53bd7af819bc24be61d984f2d9fb292c4231d15387b9ed0c00782",994]
+["21ef8221a86b18b9ac280a05a839b77af76f35e72b061dceb686de65d5df7e98",1186]
+["b6adfbe4e2affc663659862bb5ccea15a825f276666a9ca3bd3c282eecdca305",994]
+["992942c8e740bfdd74492aea8c65cd9d6f8d61d03e31f5c90d934a6a49c680b1",1186]
+["9ee42eaa95e0cfaf471de57aeccf2ab47bf0d4a712d1da6932d8ee0ad109f5ce",994]
+["83f207efc8afc50ae8805054d9fb7e7bccf5d2e759f1f28056c9403767d023c9",1186]
+["ae2a7fdd17c3348d2e91773fff8c86b373191c8c5e10ec3df6895b595a0555d3",994]
+["60b44d8182ed94d2c80650903db843ecd854808809afa6858cb5555242b46d6a",1186]
+["b5b9143e86be7c6737bf5dc83bb20929a8ccd005d1f549b69c34ab4f5395773e",994]
+["c73f48c1251b261a36da3ca019e9bfd5cb35e55a286be22ab1e4dbe0dd992554",1186]
+["54e74ce78cd67fa9035156298d3440a5f50bc601ea3f3fc430d8a2f711136960",994]
+["9325338b1e7f26ef76aa3b66a8ce65174ade2be27e7dfe82848bec4100e2b795",1186]
+["8c5393ade37148e6a4fe28d191c16a1b8261d6b4a6e5766442891e6a80763e94",994]
+["e368c837a3b03ab36fcac3657d0991bdcb22fe1eea760fcea5cff1e352f5bb9e",1186]
+["45a82364b515b955b3bbba9e976a331af6b8ff062cbb96a453b88770f04ab64d",994]
+["39571d111802f9d3ca0198ffe7c5b56befbcca43731e39c621388ddafd5ddb9e",1186]
+["39e6bd986f783ecdabb89a0eff8ee13565eebdcfc1f0531244e40bd0dc4ddaf1",994]
+["6962eafca5d2c3f695526d241ecd32a5d0f00d50cf7c9cd12b1ae4c8b2ba21bb",1186]
+["838ae1ff5975ab499e4652ed87d7bf727a691e9f146513c6ed31d2785f67f3b1",994]
+["70e80dd93947d59e29d379d2c35c7c9156b27adced03da159ae5bf45ec9cfef2",1186]
+["3a1e8b2c8ab829a195b2fbba469cdccc2078e7d5459de0319b16e9afd1722ca5",994]
+["39b8b1b2d3eb3e0af0e0dca7814c699f00e38c9fa0a27eaece45d485e3581090",1186]
+["0641690a8af6568a4113fb601a7ea3535bb19d60a9dc1d478fbbbc6a03d07b2e",955]
+["4ee0a010af4e9849f3e13410c0b148a9356ab1ab5b7e4c0d29bc3a2197575d52",1202]
+["3def67226eb3f1fd983142c2376229662be5f4403cf30fdd8ed9273bf5a150eb",971]
+["2ae5332d087ab4bbda8fbc63a1ebc42e2e304f67651c7ebee4dbca9e126fd979",1163]
+["d3d21414fcd1ebe52ea60017ca0ee36f2a4e4b58fabd0d75a404f5ea2c52a990",971]
+["4ca201bf2661e137b80ccec15e10b61c5416403452960485f95eee162ce79f0f",1163]
+["c6b326778aa6dc1ca5fee8fcd2d534f26851758f8c6bad4c9ec0036c5badfae8",971]
+["e7154e3ca4b8579547e2da7788eb73463ec4c7a6002356fe89d556577f7b118a",1163]
+["46bbfac94ef9c062958c2dcadc161d5737682c7297f65aa2e948d1a38e62115e",971]
+["2e368bdbb533083df74b0e7bbb1b7b7d0e1781d3a6581cc6b2a206e7ae5afa8c",1163]
+["09cbd6943fc71fe89e1618acfd2bda82e69e9fc730f4009223a7c5b9c4ae27c1",971]
+["a7a6d2ae1d1976adc41896bc55464d3548d9d6e8b06a999b5d57b93c304b6a8d",1163]
+["edd21ef15ae9f9636f74c005d25487e8596b6445c856befa7c347b2720178a91",971]
+["6b97028eb32f75f91ef179bcb26db3378555175f9865be190044a6b7f8777289",1163]
+["6b0671c0a21ef94a1a3c5606ba14cef686cdeb511245bad422eda735b721ae4d",971]
+["feb577573e398a371bfe2f9e5e23ffe9415ee4ba710510e4bc989b6411e2e1cf",1163]
+["a016d1266fe18aedb1ad0260d35a7c4d43bfcc69d9cddb3db1ed6dd1cdf5d3d8",971]
+["679b062e51a3e41e8ce7392ff88266c316f82f1032716fa32fc95274ce469855",1163]
+["3b9bd8faa0e624f27ae27d2112a23d202a61735452efec122158ddc790539147",1051]
+["fdc131d24049915ae4733a16aee83d6edcdda82658628cdc71e4925b74b0fea1",1503]
+["10e420797ceefb346d9ae24a1f89f434835f1e1f15ebbb6e3a94e9db4a6c66eb",1051]
+["d4c2b9650fb567cf1ae6ba4c11cf3689cd2931df960c604aea2555d9a83cae72",1503]
+["5797fdd29482e0bfde8b63a6d9cb0146b5cbda69ae45aaa28d1a667870ffab44",1051]
+["98807145f5c0641f681828ff209730b52e1f517b6c03a097f8ac5a0b6ce685bc",1503]
+["01ba08b21c8670aef2e1ddd507e5a5093d4e13634cfd9ce174ef49da3c70d551",1051]
+["1a4104994d17ab154fd86928b8e5a3f1569a7ee696886a5dbb87142a41e961bb",1503]
+["0c7cebbf2ece5081b2a327a2264d849c46308e64a3bc76008d1b1010d2cb43b2",1051]
+["75f43cdff20a9d1a083c3bd247317545815e3fa421add09108628e05faee35fd",1503]
+["77c10fddcf30cfed80534c4b97716c8cabc8206831f20b2dbd582b9af6892263",1051]
+["d025b83157695d6c1b7e3ef34406460cd7c3cca4843220e6a916541a89381003",1503]
+["f3a02fd7dffa86c47006451bbd8ec4ba50b3e04dc4acdeb4dc5e38a373d22c7b",1051]
+["987a25af0ae895a5bc50418fd3ff75386f367c22f024a1098b9079c21f471295",1503]
+["df4764966706958715d95630f2d8e3510b6f75a246a260969ad0ba17d32d25b5",1051]
+["71cf87855017ba331e176938e1047ad6978bbe0d80f5d112a6364fbe712b8bca",1503]
+["5b33f196fdec830226808151d87413dd6d6eebcb82fdf24a052ab6f4f561093e",1028]
+["004459f924e79cd89b206e8ca6c19bcd7ce0ad01ea7b738a9310cd53fba40a98",1330]
+["2d0ed910b3f5128f73ff8d2d654c2ffa7cc10ee8087cf10ab1df40720cc7c3e1",1028]
+["ffe1d481c04b67c3767681abc65b69dfa6fb90590e1f3530513f755f5fadb7f5",1330]
+["d116da7b97ea881865a06760ca7972a5a51f2513e7df10c7ebcdd45456ee744e",1028]
+["0dc1b2741bcee04f2066d573d0ca46013220f4f61e2f2555de766778c7f1e2fb",1330]
+["72a139dbeb116f7c3aee718a4ae6e00ff3cfe1ec32838b8530488cb224cd0f73",1028]
+["7d5e2e101a6a6fd5d8be914ecce8aa45adcdce63409daf27c867ddeccc9a4324",1330]
+["8b9705103ed13a7434bcbce42d3fad7e63604ca75c209e53e89d5773db18fa11",1028]
+["fd12842d56b8857a6065c311b0f0c5d6bf89c0a5783e31456079eb378ea6ae59",1330]
+["746d06280fcf0d1d39cdce73b56d32f7806c6d2a61c31edce38d1bed804d2f77",1028]
+["5dd063715e0fc202d98e12576acb58b7fd1cb55fe4dc39e8f4525b790e6af499",1330]
+["4b1e0af9d0cdb5f1a519b32698946df2255fb369a3b47b945067c0c074b5028e",1028]
+["05426fda4601ff193d10d0abf5045724bd11e1d17b5278bcca9497763c0c5050",1330]
+["da6c302428826070aedb5151e45f9d260c7d33708a93a0172cc10ebd8f918c28",1028]
+["f0aa8d049b2a022dca21e19a4b67029fd3a726b72f3652dfa482761843932b7c",1330]
+["48b13057795398939ab6d63f2e664d6e4fa44a9c4ccf385a1bb2c891372b501d",1108]
+["551d8a13c4d411c30131e72f4fa66f74a0eff9e1717d0469273510954722696b",1704]
+["28e7eb37fb233441e6fb7e5f3d2a6e58800bc449a9147aa8f4759616e988e52c",1108]
+["0e9c78bf327c7e6fe7c5d325996d30049fc77c2f46817c9b3aba28de5d8ec2e2",1704]
+["ff245484d69373fb0c8efe2bd7d4367c962909b7ed8a3c3296dcd39884f9bc7f",1108]
+["02b748b4181f6e6ecf514a502a0e2e41340470d8c78ec5f95768c571b2fda4e0",1704]
+["5fb5a83d577e9062a6741db29adf08ed774d6c18d60c2fd85d78db9ac2c38a5e",1108]
+["582a9f16a744359fe836e6718526840fce419a070becadc7b707936fd77a8d2a",1704]
+["973e3e5ccd162b5c0396381927f537a21c7b90311439566d0fed3eac98416e9e",1108]
+["cee87b801b96d4512bfdd7dbe109959d0e7c2c726ef0a6e7f474f82574988655",1704]
+["038c431913a4172f8a3942238db59d2281db1f661ee5cd4dedd34215a97d4be1",1108]
+["5ecf6fe0218be775a4307c0cd12a57cdbcf360df750c0be661b296ad5410100a",1704]
+["c42bc6106b14f8ba209e114e5aa27a5db0285303349295f939d2902f5a16e68c",1108]
+["eb0c8ee7f4343e41d0f83da4d8cb6e0bf012bd123bdbed0445eb78f1c1eb61fd",1704]
+["42604bd2fd363d0b61872ab811051824d5fe85e0aa0c1dc521c215ce1a66ed70",1108]
+["2da836980dc4b472b0fe100fff67d26e393895a28c3603f329b7d7d9b14be009",1704]
+["9652be8fe8890378412d5d70c909a8e98586de0504e521d7ff0e60d008a1fa6e",1030]
+["ffcdc0e4d1d0f9a11f5ea472027799cec92593bd9453219f7dd71432e59b67b9",1330]
+["c4780d6598caae5da76218e6c9b8e5cd2ba29c97177ddcef67c5b2782729eda7",1030]
+["ec6fa0bc0dcae13d7e17596297da0ea3df6de5298279c787d20c5048a5db9458",1330]
+["7826b75aa467150698aa81ba999598d3b31915a1e2c9168c1fb3cd1cb12c365e",1030]
+["246cbcdf5edc0e558d29d99f8509bf8b4c4aedd8553b90fe63f00545d9a2df8e",1330]
+["43ca55773da1aaf4ab565a6603f506f2f445455cdd5137baaf9f4f1c49901aee",1030]
+["1272e88697bbeaa9a46fa6391042d3ddab8b78470b6fc7f58d5251826f81f4ec",1330]
+["6cccf7375bf81a4a096fe3523f13251420f25ea53df8590181e28ec8d5b39e96",1030]
+["a90fcbe50b32b7f2b1728e7b6962e8b160d599dd62cbee50ee7384c37669e671",1330]
+["bc47dd30ffb2602e4ca07ed9f0f5cb3abf03ad76e4016311f4d0ffb94c8753c2",1030]
+["1560d42cc83079c40bb59fd5b863a12b0d7bffa876c288ff675d57b97fe3e428",1330]
+["7840a8953212bb08edf6447340c0955b1cc221696b8bc5fbc9f19bdc89dc4a73",1030]
+["f032d5fce47655d82d40684f9cbf0e93f16afb17c55d6cc977adaace97c5759a",1330]
+["ce72cb0232cd25a5e7752a69ef82dbd1dfae40220afcca100ad3fbc7ef5cc5dd",1030]
+["ac9fd223a82d92df65fa3154def0b652bc2e47bb6f1eb3f796efd53ed4d971fb",1330]
+["54378115dde366fb153aad22a54b41ee9f2690045309ccf130e96841c4eeb224",1110]
+["61b7c81f4c3821a97f3fd46e76408ed2f8922fe495327ebf987878b65ddd7f00",1704]
+["6583862a28a1fd081d28c58c119feb4d64cd24f57620b545f31c35a8c1308fd7",1110]
+["d75c25d7321c1eec702f7b5d28358904726648f79b7aff6d1332ac1bb01aa38c",1704]
+["f12eeb803eafb270d6f30378d7109de87627d72209f358177b42666db4b591f2",1110]
+["020a068583575ed5c42051d5d02eb7ab9a432a7b56d3bf180402b6294b85bd5e",1704]
+["ac46cf52e7b85d47ab80ae959e424c4a89992c615894394c6ced51f7ce9a7227",1110]
+["2b4ef31a5a3d40b27602cbde9eed5be7311f5f6e7f9cb781c405605bf8d4a79a",1704]
+["560e90c736d9ebb782c06d21aa028d03319c6bf7dddb218cbdb18728197a6cdb",1110]
+["ec9ed81e311ff5123caf116a4f676e58159abaab8d15ba9b3bb17ff09117fdb0",1704]
+["1c325894c541c387e0255cb41c4fb453d3625834580506cc2ecb9abebd028719",1110]
+["518708dc1ab75b482544d5d62b8844874b174b25694735a9fb56c16b6e7ac909",1704]
+["20146c92a6611f659bd9d48fd9d3912537c3fd954c62faefd25bd159792414b5",1110]
+["bc1e0460fa2f82f63db5274f35cd1f41e8d38025010baef6472ad604bb598f1f",1704]
+["6af38e142ba9a77be5bc0942e8d1d2678220ca4f33377a3007b01337b510c7ef",1110]
+["fc8217cd375e4ed6a7152857ca2159c678c5b018cad087db3b9a7f1604f7c215",1704]
+["24689e2d776f696a67709ea3213b49f5f13509c1a3c269d82558dc4c9cd632d4",973]
+["7f416e09fa610d12fdd4c930a9b86969cfa1da3e846dc04d9ec80b5140a0acfd",1163]
+["960fb42e6e771ec688af351eb09fc5524e1e3f62c15a6ea85b01840b909b1c91",973]
+["2a5389df02269e9a10b62394458d78fca1ccfbdbca34235312f6598dbc6c0822",1163]
+["0b4e6235de7d061f7c2d8ac7929fb5492f150dda9d9f7ffe6af64af6205c2668",973]
+["d3d7e0aa794864744c552515489f1fa7a8b41c73c6bf1ee907fb8e756a3418a1",1163]
+["3eaa51d7aa858075fa67cbbad4beeeff23c51e003ea3722f6ab9287b463c97c7",973]
+["4cddc65f06bf7f844c1b39d5e260434c8c0ef9096745453abc6f3e018394021f",1163]
+["4d53dde2b29e8fdba64747f946db92b17c9dd59e80d4592e2e7da186a884f24b",973]
+["720b8d9c56a4a089e7bcb2aebed147bc60a0b4f9d2f6425a0c563b867a59913a",1163]
+["20e3c26c0d93583ebccb7aa87215c651fcee5ee1cc7fdc3ffd65637634300676",973]
+["dfe651a9b6df0e2689979fdd419c75ba3059aad326b832b36e993ec483fd7ae6",1163]
+["25c3740d901600fa6faace79cb4cf5a365b7d503140b2ffbb0f9657268e9f5c7",973]
+["cbb05085ac99eb2885807725d5a61e34427e2068b283efd98c9088e1cf64057b",1163]
+["fb0cbb82ad86dc60b5d9f7ed2b8c2959885a5a31decec3a5287304050fbf8d66",973]
+["15633f84f46a3d14aa308a81ed818b7c39d751971af427b266da8c32b6a53210",1163]
+["df24f469f809b7cd2ee9d616bbcf81b952105bf1efef5c54ac04e79359268d7c",1053]
+["54bdbd2ac587f747af477f32c39caa418bc582c2fd08ed9ae4f54394211f63b4",1503]
+["70c2f67c3f4d74c474124bbf9965ea34c59261da55e9fa58916d8fdec091aca2",1053]
+["08545135ca087d83c852b53583ba15e83e50653ac1ed06d51dac937279fdd624",1503]
+["5bf0ff30347f17b247e4b18a023cc6d09dd336fea590548296e9c54d1170b4f6",1053]
+["38dce3d991854eef52e96a5d9c21f25412ea4eaa9819e46bc2daf545c1d88492",1503]
+["8767c1d50881971ba925d7d62e92188cc9b3a8cbc7bc0102ac4ca2c8d8775063",1053]
+["69536086fcd25982df6ff5fc9a35791d5ffb7ec0b85f7f55a0fa4c309cff4e55",1503]
+["40403440481eb0035b5c101559bea659fd2084f031400201f00e76455d83ca84",1053]
+["893c7d225a7c82d5839c5ef2a8e2da96f6d18353fd7d7db61b391b7cd80ea8df",1503]
+["155a40e1863dfec85ed203fdbb9cbc41c7e03cf6a59d8f4cec0d1ad073f97552",1053]
+["abcd22a20feacfae865ac49c09aeb32d78f480838dd19775933d018265118f6d",1503]
+["00da5afcc60f1fcd77906e361a4d4d0aa58a05354bf38de3429922324ced2899",1053]
+["bdb3cea17973d211123d7c4238b004fbae1797785620eeadae53d606574547cf",1503]
+["abac604cc689bd99d2f23f680a159e20bdac1a6022bec153fd607971562ce3b0",1053]
+["20eb36b494e863f96f4a055ea76597adc5e7d03a108223233da8b4cecaadd523",1503]
+["3c409c4a4c551e3f94a3a574783759356cff72ba7cbab891be4abbbe3777a50c",957]
+["1ec230dd14ce208d651a0894a54efbbbc89e8529203d8b38867a3917ffa6dde8",1219]
+["de387d9c40304329079ce132f6dd433a33d47357348852da6709466df0c3456d",754]
+["77d95de80f1eca8c7f14c7c95a3bc2fe94d0017fcb1d24c54c4a35c631fb7b03",516]
+["031c45335592ef58e5fbc4a78e877fca6e95bad8cee35990c00ed3b728dbc974",758]
+["53cb7c8b51481df9d5dff607963001931327b3eb604a0534fbe337500807f386",532]
+["6f66cd0bd856eb2d6cc5942cd13750b95f08ba36ee594b37f66719f905ddc231",731]
+["76e98f9e889b5080b6453a09151dd6473ffb4ec3340634bad1945117268598d9",565]
+["476b6629249d7a55570d73dd73dac37ee107678d3fe8f75dc8f67605fb9d98cd",735]
+["35f457c4966bfec5f57ac75ee63ac0aa8ab4e862e506c7c3080c07cde2513ffb",589]
+["7d2d9e7d4b9748dc43158dfe0fcb990fa39bdb3c130e0ce20f08f4651d772aaf",1565]
+["d5c44da0682f5db508d52400d593989aae0a7a1e0d31808d2adcc93a6ed94a48",4655]
+["77eb6d76448f2bd2990d064f4be1e691f2af9da9409d03f85b73a5b4e7010edc",1582]
+["191d48921b53ad85a7253a4720263d27d2109ba294aa9e7cbeeb20fe3f17ce9a",5017]
+["bccb660543d35f2ee80b404d0b584d22a5b88a5faf8bc2a3ebda0f4b04ec73f8",805]
+["f5150c7bd8258eb67bc9cfb7e90e581c1bf0dc1b17928ba9fa21e763bd777739",859]
+["1b1728b0e136075bf16ccbdf0ca8660bc63d74e1a45efd9df22b454081b96cc0",822]
+["00e48cddb888191a37c8d3759f24aad11fbda0a4606cef01970b28a4c3d14251",921]
+["31b68d23d03719a4e0aad4e2e21120cc8442960b06c43d216a5da0a8216f91ed",851]
+["c97a326bd49e32f533ebe405e67459f2008b8adf0bddea128b421d4421885c16",1101]
+["e9589a55b4763c929ca10402c080080d6b1ab960e75c5b7b3a4aabad2c828d6f",868]
+["c914e8917f604985779d75fdc212c3a531ee2f23fbf5f78b739042c69586b6ed",1183]
+["74a3ba2bef5821dadff89fd9910deb7eeaa936118e5f63cb65482d4350ff08a4",2397]
+["6d07006f8d0b7a8eac5be75f1c16760835ed49c7b5bf400af69726abf19b4358",8719]
+["112029297171395661ddd86247c344fab66d7353d9facb88cbdedfaab1edc774",2414]
+["ca8d5bb20bb5e4c7c359ff85c87d7a8499250a00627ea60f3f875dc26b55a1a4",9401]
+["4385977ac203ab082acc78033cb759e5ffb9483b314afa4f0118552ca1e2456d",943]
+["b8ff0a90aa489f94bb708a06fc8c8e21038c3385a7e19ddac950a352ddfe737d",1586]
+["af953a4d8035419e097d1df0b2c6f65e5ce6e6d15934be88f6f34b4357d5e089",960]
+["c0704a215f4248d5eab40ec91e42116c53fb5dafba81c8fa0211d33d6840d44d",1708]
+["8a52561f23f62d5f055e99b40ca08c34363e7cdb85895027273c99649d97270c",4146]
+["7216317e4dfc26dbd5681bef41b9d0f87dc58c8e589bdfcd469b47e58e043648",16966]
+["991f71ac4c15a8b8ab92b8d717d0e7cb640bf26255a59941ffd2aabc8811a842",4163]
+["0b3eee623f598c41e121c916f15902926704fc83feae6f39cc3202c8d72c3024",18288]
+["66b2b5e4be6c711e941ee5e66a2a780931d9e539de49535ecb3640e9c9b67c22",1146]
+["2b4fb666d2dbf3d922b6d206051b4e0075f2190e5273e4a3db43d50747b9aebd",2584]
+["d563ef7c1c04311bfd98cfc88c181f40a69afced173e5a3b75b4f86b7cf243d1",1163]
+["374f9b929823dca3651683ab10437ddefa7ae69773c577ae2a2491c5da2350aa",2786]
+["310982cea75aa25b3eccde0b65ca6a8c57b753549d0bb96ac9e972a56beb9115",2397]
+["f370667469c2e22c62428465fb9a99cdb547f15711b456c7e86d7d99b6db6a7c",8719]
+["6bcd1db3aef88c52b99e1c6b25f6356c3cfbbd424ce4b73ddddb3e647f3060bb",2414]
+["5e6d43c8881cce8301d2b20e91b0bc69fe1056cb73e3bde8cf867dc6eb9446f9",9401]
+["56ea691062a569936e2a11fd42edd05da69b111c9382f1c2c78a784b71228ede",851]
+["dc62d9601a186a7503e75112b28855ff25e65c5748c44147fc3fe75f44520c3a",1101]
+["3e5942dbd1b6bcfa47c2f71d72fc0af2f6cc039af5a9fa50bc46cdfe8f42bcf4",868]
+["76cb9d90cdef458a54c0681bca0e0b7a1a4ad5ed9100254cb6543b18e9b0ebb1",1183]
+["bfc2f307e83f6b7eae869e4971222cbbb314a2e2b6932159fcac235f580d3867",943]
+["9508b6d57576262b29967c6871850905b3fee7a1bc0c0e7fab48b39a8b1e6889",1586]
+["aa821875384352c63080d1d3dcbcd21730e707f656ae58785e525e678d2800ee",960]
+["071d2c8bd3f92ae8e974b6df1b95a3afb8734d841dcb9d4b97d134903ba685c3",1708]
+["ef85671bd446046f6c6055a49eafb3abb9e55035fd3207a34675b8e27f2769cf",4146]
+["6ef99a6c33da20218a119bf6f62a485400362c282e6f5f88923c7f0e2f3536a5",16966]
+["0fa720bb0024434e06d078f508dbf53d2ba1f46b5ce2a763d80457f6b2dc3008",4163]
+["a8c06b8faaadabbb49185b1fb989fc15ea1f6544c287f50d015b6e807409d4d9",18288]
+["b26e261082c8f3ed6bb265270f9c99037b95d0cff74a28dbf4686ad58470636b",1146]
+["c0d998854f196676dc9f5c601927da4bcaf0f333efbe18ab795820153ce9dde6",2584]
+["7e471596354e039dbbe8c8a2ee6ecbb525b9bf52812476f8c9fb9cf4cc1ec6c7",1163]
+["5f7aaef00f5577df32847f162a4938ffe4d5bfc86fb188681d0ab69d96ee1402",2786]
+["95e5d1a41e636f1bec3228f724bdb273e120fc0c1c58476c73e33f9cd0272469",1562]
+["4527f34c0c45bedc3750b7277159ce6715a19f47e46eecf8f8a845ef79206023",4584]
+["24ffe2de3ad20c102fc4889e88c2a26cda281c2bf6b4adc97940bd77e6f9da3a",1579]
+["eff200a4138305c23821883a4fe9d31a9a165d4d7267ff0a3cfdbe67ce370400",4946]
+["43574e6a767ed5e815189634f3bb410077ca7646cb672352e16cccc73fe46995",4168]
+["1518bda6be99095ae07e04161e2114ff3c573dee299f2b3cabba447b8d9b1558",17496]
+["3a57fe3c9c7e0b7dfb9a41df89b34eacfd811f8fac45ddebd1489594dcdf3957",4185]
+["c0c891eda156680dc8513e76b2e7ae35c22b5c1a36d3090b152704d042ca93a7",18818]
+["8e4b10e0dbd349783d26025a300b2184015447d6cd20bbb7f09ec8d4fde5224b",1167]
+["c2c8ccad75f5ec5e53e1f3ea7c047e1a49502cb75381d24b091b4bc28df859b7",2666]
+["64fa6dc33cae342f70be3edeab1a841318b6ce218fefd9934a9b26e0694c6752",1184]
+["8de8b04dbd56aca663cda0f61f0b72befc58f3316e5168f63dfdf152892a1ee6",2868]
+["077bac7412d71b31c2c41731d5b8dce92868e6c67fd3192e75e93a42c292de82",799]
+["4e4de278b5316ebccc70efa5229f1affb966e238a61704104c1e14aebb6b6c7f",751]
+["b0a0a70881024e48604ec98aecd286b5948b78ebbcddbb07837f1a95d84d820c",816]
+["a82c78a6255558145e6dcde5ba0e9b891457af06feb16a748c81361ce960f7ba",803]
+["8ef520c4fc78fef413f055a3e05061be7feb02bb64b5b054d94a45de8f7e5904",822]
+["1dc112d8a7db5eeffe574323f86edcfd649b2566cae4705a0a90c97b6e7b7b2d",874]
+["7a5085ca0dbf908d222ad1e52512d89e3c757880b5ab7ff97ee57806298f0b25",839]
+["a84277b4aeb23c68cb3604dac8c65c32eede1cc7b48cf6ec9abb4d070fffb1cd",936]
+["68b83ded5bc3836d23c159b78676a6666a7650d4bc2179f613a873fc65f8a97e",1583]
+["ff919b39989e002da2b353ab04f3855eac73ffc6a3b1d6ed434dbff75a104b38",4730]
+["35e8b362bc090b99bc96342519df73d8f9e9b665897b941125ed4d592402cc95",1600]
+["a5105e9a345d4ee068980e34d817a6d31fcd0c1fce81e841b418662735731994",5092]
+["af72b734921673070f9ec9f5dbea190757d1c8f25367e97e1b6821240d4dcca2",868]
+["3b16e7029163ca82d4eaab8e42d7385efda3167704c2c64dd8bdd1d8ff507218",1120]
+["decb9aa88d12ae172042cbb1c646ba7b03374a25ef558e96a7e794d3a85a3926",885]
+["3751bc6b08b2a92232e94f61c150a6eaee5b3148a8dd189b9a7e2895810a9e6e",1202]
+["3948b09f1e98d03b9377b27ed86aefdf262bfd411ed7c6575b6c5082c14ec9f9",2415]
+["ceff825191e41673addfcafa3c21db7b0e427360b0d926ac4abb3ed159293d40",8858]
+["1131eb682747f8234a84023d14ef45de76a16344672606cd1a2f3790fb708b86",2432]
+["50cfb934629d49d05282af7d9069819ae42d81074b6b3409b4915945de174403",9540]
+["b4db943616c37fd020960dc27f3bc8c826473cb83cff3b521c13003a3b3f3d1b",960]
+["3d0ceabc68b32fdec91d207f8d6cd44649388ca01a9b6876ad0fcf79128bf9d9",1613]
+["9f018f76517f169553c7c7b4fa3ae766180d57a11836133f55b321ea4cd3cea2",977]
+["298792cfeb5bdb00700e81e5e23dbe2f160d09688ba8a5feb89c4f28cff0addd",1735]
+["7d486aca63e5117c13a5575c0a52ac419bb0e9a9ec4c655b4e0278071c882ffe",4146]
+["3c1610dcc140c4f619b5f0df53efb49e6f61fd211506669271a7fd5504925c89",16966]
+["a617e273ed0b6d2cac003ffe6dc0b448468b3a7e44e7ac865d125b0d65c5c301",4163]
+["6962ba667cfda02bbb66c5f18b4c69f8165b5d3adbc84b9e2004c3a1b40e6fa5",18288]
+["20af765c5de893aecb99e5370874320e555e5181d66f312669c6569de6463e99",1146]
+["1a5b41dfbf8cde415dba717d8d11453a7c61a3f812def9031fcd2f359393a878",2584]
+["f42635ee197a111919542f6d5ff3d0aa501215aaed15083fced5e240a7f229da",1163]
+["7a676afa24964c38245c10df4c23e619e8f816d8a1f2bcbd533b625a224f08d7",2786]
+["c4d1dc4c07d818ea0a51dea9bad28ca93ee0648340ff390343c1cc8e2347a1b8",779]
+["6d3cfaaadf2736545c4450b19ef3ae76be4113ce55fa446f7120cc2f41ea7d4c",729]
+["66fd0582287ab7930748d0d9a527ab8db9cf6f5ba440d508ce7ceb0f4d3f01b6",796]
+["5700b9e1b4cd280589c385f2d5e4c019cd4dd044cfb288b3001087335b52d5bf",781]
+["f0b2812ef19e80b7d61291cbe0483fa9ba1fcec108c2dd3b8d522b4b2008246b",802]
+["046a0f15068e8a5dfa6586c844b8aeeee25ec1c9748873cf03acaba63543ee30",848]
+["094241310bd847092457d71cb6fc38cb1e743f1e6d21f50bec7def32964ce8d7",819]
+["c91dacc064118d22c75586613dbc3a47888633c1122959dce6feca0629249117",910]
+["8dec4251fa578b8ed0d78119b811660b8afac404d2f1d696254fa31c386b7cf7",1562]
+["41aa39c28cf2ee1f4d647062f22f944d805cfc4421c43f9d12df32c3d1269d97",4584]
+["1c42b55def8e241b99b6acf109bd9cf4a933e40bb1e182f47efc165542e7dfe0",1579]
+["4ea3599be495a1a9058f3093259e4b30a574e2c9ecc38567dd41e26bc8753ea6",4946]
+["de4c2e769dad13676a79afc24e8e5fa0eb809ebee306e4e51f3d9592d5ebbea4",848]
+["d907cd28057ceba9fb51ae85bf31afa914d374946e418ca0cea72951f4188f7f",1086]
+["64befc71a6fa2e71bb6b114313e361f39868ee9ac93de46824d8ec17b3e24998",865]
+["8592e2196b02da49b56fd53858e1a92f9829093ca049273b761bc78781098106",1168]
+["211dbf573aedc02b3dc13b2f0a092a6f003696a07717a4f5be0fc73fde9b3408",2394]
+["246bd390948591f8e332e952ee2b0b9594b3b6f23d3f1104ac4a4a238f8e6a8a",8584]
+["c5c81f28259f897d28e2f27fbc393cbcd9879a0e12c1b0797e85a5ac0ca7bae1",2411]
+["7a448100922d600fcc8ab89a0668187afc4fe394e30b0581e5ded30bb87d95b0",9266]
+["44c2dce6c5b8c392637ce22beafe95172eb8e3256d2ab9407648223e33c99910",940]
+["a5fd304a39ff1cc87aba6e71935c75739b823edb875be59fd019fab561902433",1563]
+["94371d92977825803747e8a11beffdc9fd88667e5220a6c6ca5b6648758ff683",957]
+["f920d04c14518d5e9270d2dd3e5049dec532a31e79323ed40ae8f0aed2047940",1685]
+["fca77822ce053593664d53152d28d46e182533719ecb7587d2a32dcd63ac1c7e",4146]
+["8644d3fd34645c2148126fe9661e84bfe8c6b5c49a6df043aadeaa2d9073841f",16966]
+["b96dca0493e9c15f1e777a60a442c484ab22a6153526cbf9dc04233b4a712ddd",4163]
+["246322aeb91e62e908955e42ca2f1998384c164b42f134559accecfb0f7564cc",18288]
+["bec400d24a6ed0bde90a4e53d3f87a628925fe4f841d146bc68a2d6420d9f64e",1146]
+["5759bb66d540b18d264f7f36568dc3dc7c23d9c1931e44b6b21a7cb1ec9f1fb8",2584]
+["6113cbe196658d91a3ea8067196c3445a540dc3f3ef23fa85c03c8157c5c8572",1163]
+["f12f3ebfa60ababce7347394180a91dd6344a903398b729df76097a51f3c4886",2786]
+["e20c26014b5b2eb33b9088ab0310569b0d036bfcfe66005af7cdf4a2da04d472",779]
+["ab5bf7cd5d1703a3c30b523ab675b2ac275a32647695d105a23cb446512dcd42",729]
+["152018679ee572eea4e1d5587a1d1c28bcd9b909023f0ff28ca8c1b18e8afdc8",796]
+["8c553bdfac95e02dee851db34fa4ae7f04c12a656fa8f977cf5073b1a58b75a5",781]
+["d09352f8ba4d6c8f3cf1f7c2384dbc3a511e5714649902bd1cf0858790f37279",802]
+["831596ef6d87f218ae05b1dd638f9485a73d051b0ff9c6039459a8fb80311183",848]
+["f8ceb630ff7312d328f985d8f7af25bb981fb8abf52b50949d35756801daae75",819]
+["6c8b5aa3404e9b295667fa8418d8c53a7818ba1fa9990730eb84a37780d6750f",910]
+["727d4d9dd2ff3999fcb1ab57d1ed19344ced22f9cf21c3f479c83e0a7fb1d2f7",1562]
+["fae1210544b4a8a8cce0824613159c12f00d797f6c0a750ee1c8e774b1aff45c",4584]
+["b8f646f01f5adc64dd5594477a218f26acc74fc65d40de13177123808c4df0f8",1579]
+["8932ea36f43903e1fc1cfeca9d6639b1185f0227b83e389b142124942d55ef6b",4946]
+["3a0de022224659bacc4e291ab7247ba99e4d72d8a0cbe8c7db58133b65d6b12e",848]
+["9683e45fffdee9fc7bc4283ac07fa738dd255488537e9425db16d6646b775cbf",1086]
+["7577b2b7d8e8a2e7532ac9403ecb6c21fb88d734e41189f19535dfbab3b43186",865]
+["4e9c64073aae34f8982c43da226724f2a8c6ca160122439d91209fb2bccba40b",1168]
+["51bac90253add33f8df12dbc3c850277b11c0459916a2ae7b660344972810228",2394]
+["0be722ad0fbb698b077934937834fd4837b2188736bc4f09e9ec3e7ef196e47e",8584]
+["64c03995bfefe5c7417d461c8f28b742e64f67122936524750baffa17baafd85",2411]
+["38dbb41573c2ccb09d61424c43ba56c78df6a798fae2e0aa4a7948a1589f38e2",9266]
+["ed1a9c86145911956e0c498f52d45a19030747075c5e99ccf02576188cb07ea0",940]
+["c6b2919cdd5de4a2ba898999c67f4124028b614622f533de825d270c7fe498c9",1563]
+["e8febcdefbea20c47a2d759cc6c7ca40a1421420665b22b11444a2e360a8724c",957]
+["22bfc1246f58c0e680af4ba6a01ebfc6e36c199b9c6a720939ff6ff161d032d8",1685]
+["163bfe6eff6af1ca958798f4c600442c2776ad6e34545a2a376f26497d45f25e",697]
+["3aa2cff0e9054e3a2113c10084c560ba6edc3b0b1cef6fe98377691b2825d76b",436]
+["613c83999c0cdc0f49a66fb1792bd1cb28cd751224cabedef2309857b7280b22",671]
+["a1fca0a9aeaf88b072cb101779edddb56950efe4a09c2bde26af6699f7498e86",781]
+["1c5c38962b81cb9abfb8642e0500e4bc519271564fe4799781d779fe78fb1a62",1720]
+["9f20a46831cbaaeb840064a55e25f57f9d480b764b72d52d552607e9021ead50",685]
+["b97fcc14bf2ca8ebbb024bc489ff4a785dc908fe1a335f01d8392d996e828fb5",844]
+["b54ad3b0e1fb01f5f838e6e2c8f9f98727882fc20e904a73b99f93c181da0c39",639]
+["9d3a24a945d998babfbf411193a0791e3efb7d40f4179c19e40d788b4a9028b3",665]
+["209a9ba291d8bc1d0a302a96aa95c50b8d96b79c8642f8e6e6b9f0883c6d891f",637]
+["1707961af385a3ce1d6050302c8fe3491da5ad7d7c7d2082facada7fd2b41eed",648]
+["aa3bd1f2ce23204e107cf830ec3c102c3946f2bb7ca26d76e429198f265f91bd",676]
+["16998d821efa80c0c680f9fa3f1da2a812cefdc428e72e08d140f95cdd653c42",824]
+["01e45193748c37b3b4c13067dc0a5256cec5157b3b085d499f1b493cf0c38c3e",677]
+["d31eeecb4a66aedc858d6224a69e7c208bb0605df62256c5b936ddb01f79d461",808]
+["b921002eb50e59d3cabc8069d345271acac508c79e3d386dc6e024928b426722",683]
+["dbe77b3d01aabdf0b084fb950f91b087e8e18bfe8b53a95b3a663e2e5c15d956",818]
+["96a6e14bb97abef3d899dbb53d3780f02dca7f10d0b825f6f71381401432eaf4",653]
+["05711b79d884d5c5f48e8525bbbffdb1e37f41bd4256057c06acdd63f29aac1d",704]
+["516bc7651247dd15541d4f88d76646331218bad0e092f7807590af27ab315ec9",652]
+["0877cc6b1279f024e9451203da5b63617dde0d6cceb1e3748fb850da6b63c486",751]
+["1c7fb466d977d62dd753a33ee8dd8459112542615f291b85b52c9308d3c79c50",653]
+["4ee323944faf48643245e69e116f08e93e215c02e2d19840b962417bfc6c0869",699]
+["94021567bcc93c56bbc75cad94a85162ee1dd0ee2a4042a96e4e9a3f8f079d40",2248]
+["0bda69cbc72808bf93a5c4ff5f3a0e8e6b7e357a682d28b6b3469af389906312",643]
+["d62da2db2209f6950061cf8b664aa9b1194067949c4d27c9ffb34447f522589a",669]
+["f284f3aadcb0c7f6ad2e2305744a87902904ea702558d07a5a5088f281cd3646",584]
+["ea3ad60f482c2f568b0d0e4fbaf36a82f830e51ace48ae675f23348a37276776",2336]
+["cfe465f2abfa28c27c584e86e684d194807cbabe77b18b872ba83d1bfcf59175",634]
+["9a2ce5b6bce81169f4d7223b7536ac803e14b1d939d9e6a9d3968cf5c779b295",448]
+["6d15db91a69432f86447cfec8cccfd04c5e3ed24bc2516f110a7323f96d7d460",584]
+["01019dc7f5ad3badce0879b952639977ea567cb67d9d5c2470b2d768e0c9aba2",2336]
+["4ab030258ac00a58cfab9890c7905b90d0a71cfa90f1b6ec06a310df8c83fe4c",634]
+["2b969973109edfcac3743b39a7f3c6d06834147e545e5615c8ccae7d77b6160b",448]
+["7cf84fb55097c8cfc350711b212d4c1392714cd2eced2845c64c59bc58e06501",584]
+["9e2a76cfc327c0266c0deaa0f4fa1eea7810a0ca6e256ddd24b0119cc2415591",2332]
+["09c137cbfd07851a27efcb0c4fb9b22eb776c25eca93a6556fc77a37c7466569",632]
+["28ccc710bd2ffeb231b97d9032593e4cd470f31befc323b1ea2be53d0af8bd3a",448]
+["beb7751a12d07a1d0432b564da9ec1dffa1f796ced825d2e32c2c1192fc08d4b",548]
+["7c274a11d8d47eea1825f94c3b99716f5f74cf7341c0a5b63bd55ea71488bdc3",2264]
+["93343b9c449fe660c8621b321ea974010a8e30f46139c82d856c404629d94cc0",468]
+["2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5",352]
+["02faeb5e1fea902c8a719e22fdeb884d7fcad9f032286b63ff4ac0b12a575cdd",548]
+["3241aa33707da15e672b29748c2f147b0394faac2ae9ed636076437f2f59dba4",503]
+["df5f25a610d3ed85ffe41cedbff212bf195923fc627c73fe8794c0325d8fffaa",2296]
+["bd7a907fa69aeb0be694cf7b64520a5f5cccbadf2fe92b8c15bb7983147ed102",556]
+["e268b2fc78cb35ed4f7d57a84f4750c0a2d99e2cede66ebce22b727fe9a1085e",2284]
+["029c64a2a868fdde54f01ad1c5c57b9524ed1293e64ddaf6101277526e72fd4c",488]
+["6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a",371]
+["0c740d9346305c8bb28f02d75077e5d120f5e96bd5b4d1aaf5efc51651d1c697",556]
+["4502b138346f42db6df26653eca2ec7c1b87f5605230b91847446068f578213d",529]
+["0aaee12f86143cc06de9d44fd66584fc2f968a99c2ecb70f4b6fcdaeec9c2382",2312]
+["85eb7a99ff0f6fca48253a5dfe8fd7ff2b9a4ad5fe91394b207222046d669532",557]
+["de68b97d57fee2d37456a398289f8b6ae0a7ab8710cc236ac109eba2fb29bf18",2288]
+["4477a3352eb4249f39b409af6b7e18aa8306e2e4153237c6c66b3b0aa324b162",491]
+["e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3",375]
+["ac5f0aaf56a6c990189899c23baa838a40cadaac940d477e650b9f6169af28a2",557]
+["afc5aaa25671f35e35d3562cf780fc3235d845cc574496fcf50639de89f72212",533]
+["e1854ce7a6f577684d597a148edd655fe671228845eb5e2b3fc6f272257672a7",2312]
+["72f0dc543ce3522a0b54bc93199be9973198258cacc846c060ddad3779b51ce8",562]
+["1e58c917f253fc46ecb68470b70341863d1b2d2e41fbea9481a08fc56394f1dd",548]
+["5de97c5bf5f46bf8ad882c0111c22492867b35fb343485e525a83280f9de2c91",510]
+["7841cd5b8492ebb3a764ac57e79f5e5d4dd2f2a549dd85dc83492c8ea642f14b",366]
+["84ce5bea52f039d2328b354bc06cfa1b5f804d926bdc3b569ab18e0ae7970f1e",510]
+["605cc1883dac9a6075d82aa930292021d8476e1f9b6929e77c1bf169cc7949b4",772]
+["cb7a3f1337d34ef43b180fa83ec85089a3608104630c77d3c9f172703b590cb9",2440]
+["1bb36687449b45daa0568ec0e4e6582d6e8a769bcf576a46ef4a2335eda92bbc",779]
+["f65783b943c54114217d433394f00117aa092c89101aee9f13fa9ee540a51fc2",1349]
+["b3424ad4faf6ebfbe83c386f71de9b6830157d4d46e7dd962cda1517e7075211",555]
+["8e46ce38d8dd8de271f59cb0e2fd57db33206437aa7788208a1705957f7a6a3a",503]
+["dbdde8c1ebb74ea7ad47debd386af71af7f497e85e5c6220714481e999372a56",640]
+["da5eb8af8df261d1c8be5e4015ee7b285520fc2003ba3f99d15833f8bf062cc6",2328]
+["36178892d694b27d2d027578eca82b02bc498c103893cf2e8a07e82317f85655",599]
+["ce43042db467c853f37475fa41e25bec8748020a6f3d65a8ba8ab6e66f0fa3de",397]
+["b1be9f32e32b91a89ce09770f776f01ca4cff5bb7641c182a0890bb3afebc091",619]
+["8e414cac8b77e5f74a06db281d7f30552462c3b8056f32c6eb5398194185364b",2300]
+["19f282f4617fc8412c29c4d4ca19ae261719a2bcc63167a64e5f114a3a03312b",564]
+["714c36f2cfbd16069d2d1aa37e70e37d7851e288255506b534fa7dddba9d81a1",368]
+["e7d3c61c24fb0504d134a4b03f425aa4f0945ffc8a71dbc24b7b458342db979d",649]
+["0513c3322104c49f6b014b38cd43f3a2c0d146eb381a79258b1f379bb87e5fd3",2340]
+["b3fb33157821011b08aa5b153612146cd4a9399ab46529d47e9848706caa74a1",604]
+["1d25175d8d4e1e379e7d37328d49799fa47bbe2f3217fb46e8f1b211fae3bb23",391]
+["50e3e9c5d9fa9186c626b97a3ca65868b453fd840705c983c706c1013d540967",178]
+["34e160fe6ae268f13c74f3f2e4c137e2c0275d2d0839deb1dc1a3176a7ae2e99",2284]
+["bd08f68da34dabd5e802c53c151a26db3c3b65e5260312d324d2ca5408cd68c8",610]
+["4ea85895c28da978a5060511d78b93b5ec18ae63d1f034616db37dfcc52036c4",174]
+["a28d8c8f7b7594cc19b7a8be83c04a9b79690e74a2e9375f21c67d109ebfffdd",2268]
+["2ffe6ffd9935b2be059c61fa1f2827af2591de71aacd784b92ee45eee509a5fb",590]
+["25cfa815298de1e74f393ec19545d21bd102fee5f5d4cf5580b26e31fe271c72",178]
+["ccbebc8b7b7f66372cb97a1a0d862021f1ce40c60c653ec78f7c048b00190fb4",2284]
+["54c8d8525ab3c01cf6b10030d7f10afe4d47cc973cb1220a08e55fac462e4ff5",610]
+["df878e17129561f077cd37ad825d16b3933d9ecb0c86f51c030c44653e959137",174]
+["8a8a39ef8aae2918bb12a6555929b60bc099c4fe229ebccfcd8b4f7773386a7f",2268]
+["f4bce8a3c6bbd2b3d250ac12e8303b28fd86b4e3e146e96fa741861a33510ac1",590]
+["b97ce3547d6ffab3c79f49862ef6086892535b2e045ef71a30b43697c022baa9",178]
+["38b15b0fae0ea9b9e1a4ad5f3bf985cbf0fe848d5cbc2a72e10928515d5e2cf9",2284]
+["def2e402f8a52a7af6620eae84b981a8080d3026823909e3c5a3e43b61213f5c",610]
+["cbfe28a8d942849fac21b0253d37c0daa12f9a56c2a795f974b726c421d1b346",174]
+["6ed2aa4c3be59a8a2e0608c28d78b6b84b46ef1b42310854a038a5e3fbe84758",2268]
+["eada6c727b2fcfb2a2109918690b55cb1a6e3412e722e96ffa48e78af27dd3c0",590]
+["69d974a9b04e187a8a5995ea3ab287447ff2706cf3185711a68f19afd5244090",178]
+["da77fcc9138901534b79bea41a01abda95203ab6ad1d5cd7d4bfdeae433359c7",2284]
+["595d5b11da63d53c6f7f6617c864cd08be53c0184eefa2bb2a7d73400e4c8e4e",610]
+["32e72e3861e66cce9855388185690dd1130e2c7f1d1f01acd3dc2e3aa0db9747",174]
+["fa0d6f728401943e3b05a9326b07a4f7039b0ee485cd8a64f82cb1e98c3037c7",2268]
+["c5c7eb12c0a03545b71e503ae0069104ebc07afd2ad2518ce8d6e266ed33ea7f",590]
+["3c1ac26bc41e90bc20c698f5a1171e8280c06f8141f089fd751f71d074d4c1e0",167]
+["d0334824dd7f67aa162626bdfae0ad165b29cbb94c34749c604a1c4220714393",2264]
+["776bd36ba8aefe26223242b09ebce9a105234a0372eb765080df8d3ff643214a",557]
+["ccc95765e1c63ee48c89b2c82eab9b445591b2ffdbefcf991b968551a207d8b9",171]
+["72c18e67da4402b4e84a96ccffdbe3aafabae2ca67f25127c6075171fdb2f306",2280]
+["dc0ff5c2721b9def08dbc83dca37d858c12694fcd118416d10e52f9c7c5c2be4",577]
+["7857fd700dca5d2887754c6072901eef91bb6097ab129fd8c3932026b7ba7862",178]
+["03ef43476a4946c9cb29ccb543b63c53ba17c0c7818b9b9a32e833f0a2f9388f",2264]
+["c509228ece5428585ff48b30bb38c46dc1ff5376fa59dbfdf3f432f173a653da",590]
+["1dfb41a7a5968fc6a86ce98d3a2a56c2924596c210b5b623add36c89b6b67f0e",354]
+["799175fbf75ebe6fd5a88a01aaf1870e77d8681d36665f2aecc3cbb9890d8c53",687]
+["b82f196b8a8d6c963c92f80df51cb3e8b4a1cd69babdf3c5c0db5f3685550216",181]
+["eeeda51c077f80ebb0f8f5709f63ad3e6c190983883b7ffcb4c31d07988429c2",2276]
+["cc3bf6712eb6ce75d59cf5e3597747ea2432ba434fc6638b993dca3466794bf5",605]
+["c4f95673b022a5def1d4faa1c1819122e8f68f32db6927e2fb507f4b725f5b6f",367]
+["c2a9b440e094363a68d260b51798b285e90cd8298a3a296fb645316fd46fd024",712]
+["714c121e3484f68a611435df2c86a5f5142a2bacec729d097254ee0643cfc267",176]
+["1d79a4f3d5603f18f7eb11c9331bb77885520a145b67017c4de69b3015754e1c",2296]
+["217622fada191901ac14e53d234c2ac70d0b423324779ab3cfbf536649f3c4bf",612]
+["7e0eb44045c24c0fdfc868170ec37e5b78c41906a5eb405e8653fc076e40209f",335]
+["e4aa2437e05f1d6da7818d175e43d8258bb23d87806221c6bc1d485c51ce156b",642]
+["d9ae89d0f3303a6a61cf3c14b2e3dce671ae6a120dc2f78a568c413e4f0d3f1e",176]
+["237c0ff11f949e340790f3a0d96bdb6d8205c264fac048fde0f8516587fc1ead",2296]
+["b936dc8ff7c89e1446a7beac160c09c4c155a93ea82f0fa59a95740c0728562a",612]
+["644fd28eac7632fdf7548075164714039bc02ecd11370550ec97f675e00f897a",335]
+["387b6241b07eb0f13cb5de9d660e19aae0d1c576fcb30f77f6ed90ab5de4a0f4",642]
+["25938feebe2b654f32e7e88b97048953298824939519d5a88d81a5528d1219a7",186]
+["9523386d7cae135925e0367faef6f00c04f9073c51a05821c4e25369d63419b2",2304]
+["a78c7221dd329955f0f2357df8bdc5bd134ea094b439e5b7cc57e00506230585",634]
+["85465f4cac93f6edad8ead3d7bb1b3ce334c38cef009b959d7ef94c571ea8362",383]
+["bc4631b099b8f1f2e45542034ea02f9759ca0ae818cb74b39cdcaf77a358e475",765]
+["46d04e91af3a69ccc5e529c2a680c147b4f48a57a2cf4e285b137ff7092cecf3",181]
+["56a4c6c44d5d6f08886b381f3b0852da8f744a330fdf2e0003e9a1ba5dd42ddc",2320]
+["2eb447d312fad4a20c0199a1badb8b92e4fbe4c4b1270afb00015bd2ace2109f",637]
+["923f6e723cd09766e94161c371f631c16609c2dd0f8b7448b6315e5e2f163b20",350]
+["81721cd5392f59c2dcc269ce2cb68f0c316ea14b0df29ffeab9aa929b34fa502",677]
+["0f4d2c5941008cfdecd590081e8dd0e3f62b5af1c7de723c93a59c070cdb4ef3",181]
+["8ff8ac81ba64c22394a8850ba958f9f0e325b618ccee0d1e3d0610b0dcd31fb5",2320]
+["425ba33548346fc01a3bcbb0139ebe77313ae5c877636a83d6c911b681a96839",637]
+["32ecdb7074d4c620570db1eb58f9aab9e5a4f108eba0afc93b71968a76ca1fed",350]
+["a5d5e4f2b7d8bf303dd8b591dccbf17d934f506e29095a791747e68f4db577b0",677]
+["19918a31908f288362cf938396654ad7c957b0b166c0c0ddea9e519673a51580",183]
+["bfba21843aaf995cb8b450e942b7ed754286f5fa56236392d8f8eb9af0c36c36",2300]
+["154b421aa305b6d0c1690ac60649ff66ce6776e8eec0b2ca3683e13584cbdf67",627]
+["16eefef05a4ccb13161884e67e667d938acd057f7b8a0f569243305bf1c6fd68",372]
+["05602abbf8af0d4f47c5b1ae99811fdc280f29a1f0d98cf487315a9b6e460488",756]
+["6beca880b22482a309fe65c76838ab360b97b2afa655a4ca904c90d2684d5435",176]
+["45af27b4f783cb13f9f7e46a3bc4efcb85fb2b9aefec577ad14c747097ec9a0e",2332]
+["670544d588e170a550ff2b71dc590216e374d67144f02732d04e4c09bd7d48a3",652]
+["c155d160709cdc9f8f2d6bc1bdf349257af87b41512e1c5276dd7b65bfd30cee",338]
+["c17d380facf2421075e82b0971f0a6bdaf188679ca4b0a3b01af01fcb827d965",736]
+["40fbf832dca17ce296c726679d6035c1ce0092f0ed3cbb4cd868bff26ff9e7b0",176]
+["2a4784dee27fa2d881838d934aff0daf713621a17984f1b474937e7ee5db8332",2332]
+["b0df003783bf5a5576e881ce18342701762f6170a2d3147a7e60b3d99bc69152",652]
+["432f833c6cd3c6e67dbb1fd6f51168105a5b69d3508de79cb3f26b486b5cc0e4",338]
+["fefc45003365f04529d9e14956ad27ea85134ade9970d7c5abbcb77a75784248",736]
+["0136a283dea73fe52782d4f090d616d9deb9fc57d6491423da5ef5efe863ea7b",181]
+["07347fd8e719c0fd43789ee3645308e2bd51b4434357260a6004f7dedcf74b16",2360]
+["19dc5044e069583fea329803e7daed26af408cbab2576307c431cecc59060b50",677]
+["b55a4b34632ac10eec0ad9e0b75aafd769487451b1dda0d7ed8c5ff6e6f072cf",353]
+["b4a9ddc2dd540fb7b2b7875302aea072521a4516b3d68a3eb8c5ad65dc942fc1",771]
+["dc1abd723b4ce674f28ab1a6372e31921b7dfebdb0bffa1d6ce09d6f5d37083b",181]
+["8a49fdc7f1cc2d69ee99bab5d610da9c7be26a932dba9dc8b8a7f0d0bf62844e",2360]
+["c036e3ad1ea15ac86726658954d576d73d0ae50ede085970807809a991d9af0a",677]
+["a399f38d25c0311937ed8b4d34776f608198571d6ed5f4b6721636f9683be4aa",353]
+["fa2ffa74ccef90a0cad5f17f79945cd2e5569972ba57655881aa40d72b68190a",771]
+["b1165a162525aa49a610c57216ceac531ecfba2789c3da430c5999e3ec7743a4",181]
+["acda30120dba9877b11aef0943ceb13ad464c6bd4d5fe166d05417c3e4c32255",2280]
+["0efcada368dccc390fb9d72c7e784c4f11aaefe4ef4341e3f5fb62ecd7235021",609]
+["a5fd0fc2b3f03497abb0d77d560952b399e74e5cf130c3f308cfd5ea6743c588",368]
+["4421b174c55c82f049b694f123777239c949fa30f6dcaf46a344ae787dd1429a",720]
+["8e7ee1cdd13560b62a3c304fcfbb3cd7c364118230977528a59608d9bb12f1f4",178]
+["a3fefe69c140508cc56f1cf46b8a375dd8d85fafb369007e40afe80aaa6a3e79",2276]
+["33183daaddb7f05a5119f87e6218c1ac7b77aa57ea1c374970e3e560b80363c4",602]
+["a7d9ff8e957ef959bb0e388e656c9aa15f27bd8045ff878ecdfe469f763433a5",357]
+["dcf6d4823b2a4ad060bdec80ff99a7ffe9c8ff10b7ae091003635ac3d10df0a0",711]
+["67848ff81d453ccf2f1a3de6409bf311af4fbffbaa371b18f82b5c882a34aa65",185]
+["30c59dae57e6dfd4392fa7d61e2347a23e277f4fb50bca2f15080ac61d200415",2284]
+["a087cd13bd9be98cc05461922b2bea678dc54f85f49a0d773ed3e90265e66bd7",617]
+["110bb01ff18541a440f5f730f11b878ee453b00477c2c665851bb6d41e1e5e7e",181]
+["ac7595c49f2989c0b544286fff2903c3290af1a1c81579c3a013ea21c11b06de",2268]
+["ef767c635b23c7bdb4ea46254bfe941def1296f94f37eabcc98daabad3658b33",597]
+["af3eef3bd25b62f66aa9ff89f8a22ebf3f983091dba969c7631b5cdb78e5a7f5",185]
+["659a8a0fba8f01da20fca5d1f0d167939685d6d50880e7b58979242def162151",2284]
+["1ae16f0065ef24ef4a9fe69aab8522e6fa0e4e20d97d7ece564acd4e1d243a03",617]
+["2b6ee262da0758c395b68c92fb457d085e82d384146da123d5ced6c6d4bbf49e",181]
+["fc493f7978589c327d39085ee529db73b4db9c116be48ee9b0aa4936eb40fb03",2268]
+["1a2951c9e8fda3e0ee63d860ff9f37f47d1a029f990610834d164c79409cfd04",597]
+["524c6d62369b53784f975e5c7bc3ed26b4c7b4b84fb233cf596df4664e8e8ebe",185]
+["778d54a84520bcad567750bc38a29ccbe84df74115b403a8764c08cf120574d6",2284]
+["bc4581a98a403c8e1b624e0a315d994145ec359e767e6ac6598fe29c6b5b37a2",617]
+["dde4400e74648821596405c33fa6e84e0a9f7a37a2476ea1b3a668d1872ae9e0",181]
+["4290935c2a5f95717fbb6cab484f7dffee8f96ded11500f209142ac94af74536",2268]
+["8c60eb3ab364416375b8d6874e55ebad155f8c723ac75428884f3de5f507ca42",597]
+["7f8510821775296027e8ee227e8e1f4fa5e011f3bbfbf02ca11b20e38580b31c",185]
+["06a6606650eebd985c677f43cfc728410f32a350caefff2591e3ddceef816594",2284]
+["0981680edbf241134984b87812304384c85adddffc29eb4021877a23705546e0",617]
+["5d2a6d4b7ecc300e59287fa68ee3a6d2c206c2154a519974c4d682d8d557515f",181]
+["8d1e7671911cda883efe53fd14928a835257a819b53bfbcd42297187070a61a7",2268]
+["fdad082de29808bbf4cbb42fcef7858183f61bec2a683d33d3dfe21e65c77610",597]
+["3299e157d0f913ff8cad892f7a2969ef84a79f4f5983a939851c751caa25d887",167]
+["5b8c9dfb1ec540fb1b425226f789113c7c3440596d2a9cc73ee682f3cf550f76",2260]
+["27b5b8fa6d7c26fbe807c525070091591afd3cfcadb4d1110a9ac3bc2a839c6b",513]
+["728e12207016cd6964d6b2d159caa12beb099ef35eee4414d9fd17e66d3f8b63",171]
+["2ede5020b4340c0283c9a4054b213dca08e4829fcec57fc9a5c12930a6571e41",2276]
+["6bbea64fad65ee3cf46042eae371ce04077d3b80d9774b02bd70ca174a937d50",533]
+["e5f7b4753b7651dafe9a3465e0ae14c7cf58683549465192ee4df3aa7195f359",174]
+["8c22e24ade5c197fa37fc50b6ec68af18d1cb1e872aa633ca2f17beb281e7097",2240]
+["7c9e94b34a87dcf03fb3833bc4c1f7a1bda42884789678ae09ff642ba2ca0116",566]
+["2faaab2e685c9cbec888014251993fc23809575e523d049cf4a5a8d96a30257a",335]
+["5a2d6d20951ca91cfbfeda6964869db8573c95ca557c4ec0c190ac2f0c26cc98",673]
+["3fea065472282ef5b9a7e295e76ea73f4bf61c3c7fed9f71ddd6f02e05913037",167]
+["7ac694b9f7f8073ff95e0d86e70ffdfc2dba15c553bc6565c5ffc1fc1328d9c7",2236]
+["2cf9e1de457e445b27bf05b0cb19f4ada067deff3722026a12a54d4ae037b881",555]
+["80d39db62f331d4800a5f85f85ade14b10e23304166fa1444aaf5cfcc41179fd",312]
+["35d01a2088de7bc3863baccd594f26f20f4843370908c85f3b3f25c58cba8c92",665]
+["48424f0a5619259f7097ff1494360be6edda589dddb2611cf0eff248aac5498f",171]
+["7f50306d8abc6dc3ed7b6f89d16461be9c3da611153dd683a10529b46ab158b6",2252]
+["c4f11c86f19c399c0c1547f03727bfe490cfb3f894bb5f1895b7a0af365716d4",575]
+["dd4aca145c5b0bc05588ff8c0cc9aefa33c179d4df336f9873a1567a57511a70",324]
+["988d562d7564eadbad8795e82b3dd3539cb4d13e6a8d76749d7671835a417d0b",697]
+["061fd0547368e9b3beeae14fa3f64d62117f736f621fd71144d76d703ae33cd4",191]
+["f6a1ada0e0823cf34119a3262413ced6a59ff78e15916e6abefea128dfccba3f",2264]
+["ce2383c7215203919f1d9b977fb6dfabdd845f4d78c0877699687ac7d78c2498",603]
+["1e3641bcc6fff4cfe6084abd9e5aeff71c21a2e5e3598c66121c1e9a14ddf8ac",390]
+["2fe93a854e0c787435c22f0b88b2e0f86cabbe46868e5f5f3736c14c077d4463",706]
+["131fec866dbfe509ecf82dba278d80f98fecff471dc30897a30483c796c10b48",175]
+["fef25b1dc672cb436023759e0543b3e01a95544e4ff607f97cca896ea2648a34",2272]
+["c650934ec9f5fcd33bbaefbf771d7f8220578bea1cbf138539d0da0389ca8624",595]
+["131f29e7648e2df83a3c8d957489a2dc4544d2f9eef5041bd083c0a87462f2bc",336]
+["3ea31ecb5d72ce4d965c9be5339939541b6ea0f689d6280b16752e8ca98a7a0f",731]
+["2a031d0028c94d26848089d4c651c6399c73cddba85f5739c45ed26a48626c6e",195]
+["e327fb24921582151d4da8a01a96767e0695581a63bd5aa765d7aead4e73b113",2284]
+["41534f09a6b967c65eeb02f0bd17294c0cd04c73112289038e80210ecd0a4eed",623]
+["adba7a3c56896742c3f11556f5c80ede7ae1d6c90a10c90f028aa068957c3bec",402]
+["fe7c58512ff9526fec677ad4e140b01347f1b1b666c9b435fd5c75e4f5e5a0ff",742]
+["c74a23aae93bc52b9be7e5a494481bd5f4c5ee3906f4f14c2ca262c19de29f7d",187]
+["6b0bbd92d4f781b1b7a44f3d26613e0f4e07355adb151bea9862548b4e82f19c",2332]
+["d680a7e23d0b1c682cd755a7c68b97503292ad024c94162fd965fbbff8e3c999",655]
+["6e50542e54ff1c0f49b776fee635acb5d6fcd8a5d7897d7dee0d11371307fdbb",372]
+["098a31d4e65f792ce0c17420a8dc8f89c54ccf31d11903b644bed62b1e5da91d",839]
+["ab188796426740dafa3e442ef1a63693c89dd14ff1b3d5d8dd7e15db69f05a87",207]
+["bc2b66c70469cfce1671084e3199dc739a20fde6a84856a895eec29d435e1f6a",2344]
+["97376fd3fae392bb9041a5649b851f6944dd0ac024d09365e8cf462003655011",683]
+["a018baddfd144b043050648209669164a8d7632a3a6154ac405d476ab89be90c",438]
+["55b6554b57fa69eb67c871c9f765e22d5f9272d5126239152611bc9262bed227",850]
+["2b07c0ee3b1aced0169d7aab176db75d2d7402997ecfea8d2c9409554f02bdea",188]
+["24646676f30b4c51775b2401f07118d4eae702841207344997c4bc058e9c23e7",2332]
+["3e0b39669249c86cc9e42ec40adf9297cd29f830f6997fefce15fea1a255c7a0",656]
+["bfe82e33c244df50b135041b6e8c059f74b0093a56412eb576a328f905455dfd",378]
+["0d5e4438674104650dd34e047116dd5583442d30b771a0419af8c37f73647d33",799]
+["b5a68ce60ffdf514460d501121483e104d78a671d8eb3bfd3138024ffa0bf43b",183]
+["c2b88397d072c1bd6f07e341628f66734af7bd152c8dae52649fa986a167aa4d",2312]
+["4eea5f0a51a527ba94ded45a4ed11e1ecee2cbd754158835d466f279fae9630f",635]
+["2c25dc715d05f48fe3ac5a75e611080b4c8b70370b9fd4c62ed15bf4bc4f4844",360]
+["e42dbc72aad7c1d4e8834e9db5b2cdcc423320ea18799df803e9a88bf2651097",803]
+["28df116f29bcf1914f35f75d1b995e7243470d909799503b4749e8af72748eef",203]
+["7526fab017e47fdb981d5fee3cb72afb0f70275b981ac24f5d685ab971de4ba3",2324]
+["479661ca38db7d167abac4635c83c0e89acf90134864435bc6d42249319fc851",663]
+["ce5f9a83b9a6dd17d15df0c6eed2cbc40eb15548d9de8a6150e71353890b4665",426]
+["5ed1cbaa8920aa12132c4bffbe84c4232e43ac39701795f838688318fe1a0b26",814]
+["e09c7b0c3bb8c2f500930139389704bb96b18c5f6eca352aea8fe322f5f39f9a",184]
+["7cc0a3d5555ebe940fc415645f0fbb5a01b6c0cdb5dc7e69421dfdfb891b212b",2312]
+["39136f8e8e87e76dc6f42fa0bb14efb5a03538bf2d8e20cf31eb033b59ed2676",636]
+["944991dc27d6d8ed6f66a07d1467d13aef6f46b3dacaa8998f011e39188b4534",366]
+["e07f221a94a31e0f93dc3805a431b9bd8ecb8796a38c0940d38a69bd75cbed11",763]
+["62080f5dada9d77ecab046fdd1115116ace3a8a0b02e74357b671f6d47cd006d",178]
+["59828f02c68185028d3089229a04d0060bfdc11bed5798be0a9ad6a0e08dbc01",2260]
+["b9d37d08f809e6cf09945f0929c3e3e85e0e841d55b0e4164d55f4332929f8b7",586]
+["f4a39265b00c13f1173be19c05db71f8e5a892fa31ff3d4f0b08550bd47503cd",347]
+["73fa1c4d219798fb643894e31e2a5d087cf0464bd0afdc179af56a1952bf76ab",705]
+["255e764e9d96bed501e8ba6c224ebebc54a337e4752cc6dbe9f2a5f3332d5243",198]
+["201a97e4d40c25d76f2f7d2effe0f8cb35ba88eaa9ee78e349e0272960b7f818",2272]
+["c320d85c472830732f2ff34eb7d3ebe4c97c3a87ce6da1fc3d16b14d7edf8b5b",614]
+["85b83aca07d33a993af6636d7ddb3a5712e70c4afe0d6652d7e050427cdbe7ec",413]
+["4e507f1976fce696668c3170dcc84a2f7a4829a5e15e228ad4e6309aa93789f5",715]
+["90e09271227dc45fb5fddfe5600dfdf244dbf557d63e7310ccbc60fe3134521c",171]
+["908065411dface5a8ff037b74914ddf72ef4a04d8d2691f425d14d9a1ec5a41a",2252]
+["91eb928e1c26db5e87119a8d8add870fb65e978f99c5c416b5aed9e3cc67ac7e",575]
+["645cf1cc02c927b10877276614b29af4695fa195552b976f80ce8fb1e8b8d52f",324]
+["0ccf5e8e394698aa6800a0eece1b1c7674c550871ba1817f9f97861a49193ddc",697]
+["0cb2fc9b39e9ff79dd4b3385ff2a863ab6f985dae0ee9a2b32953c8b614bb2f6",191]
+["81b8e55635759716371be3a868118a1135a91483c4a6e8b50254111578eb9bd0",2264]
+["ace35dc4daddf9ce1a3ad9d894a1d75ea9d7f731e141a973191783761c3b6a44",603]
+["1e4a671c046aa8bb65a0947bc662fc52d8732664cd2d9f4dc6c4b714ac509464",390]
+["9cd54d0e964bb38b958818ca284f30a377810e13a18dbffc656647f96952b85c",706]
+["c570444dcaa198e43114d78acb97c48f20a553aba19829e8ef583ddc45e46702",190]
+["94ad7a6ffcdeb17e243415219564819665e715050c3e484ffa7e6102660d4312",2316]
+["299c138b1a15afae71d5f2bb7c4e00e00f95f97466b6504ad7d5aeb369fa8e68",646]
+["cf25f354df66bf7d02f1a3313bd6a0d128a314bf35eb15bf8fac4158d6276257",383]
+["21704b0d520a533e3028ec91e1460abbfd4d10871e9868564301d0f8de119605",812]
+["1873665996ae9d394cc5337f15ba4cdc775b9ef6f114a613b412f9e3fa9e4569",210]
+["4928582d00bbd0b10c87507a323693eea18c1339f1df8d8e0887783561f77f3f",2328]
+["9af5838b3918033006a244427a7a05edba13b5b9ba85ffdfaba382f995869617",674]
+["75e58d79b8a18a6a7361fd250bcebe8deb6634b90d969ef8ba3e96fa9d818e0b",449]
+["e72c7f830a09f5f584b67228ffc9585a55feaa3d3568106bd86a2b51e14738b7",823]
+["ccbbb35d84cfc273d7dc22c9cbc4728b27c9938d429391f6783587cae3b53cf0",183]
+["be68f121b4e41cca24caa8e06318b3400738b6a80baba50fbfe67a377b05c592",2312]
+["528d4200721b7b74228dee2731937a0e0b42a259b309b4db1e56e60ada76c28f",635]
+["ff3ec8a087f69945c6cf7adbf0d3e06873ccde67357561145682949bd9b0f601",360]
+["dc7219a37cad6ed171940117dcd2ec3a715a4fdfaad7aea0ce76805eef380a1f",803]
+["e6fa5ab5895efbfb96a2d6fedbef6dcfcd137f708abefa4ae16bd30aee14b0dd",203]
+["25379393c8f866404c3fd0ddb0d6f7e95d59dea4ff199d30b867bb16545b44f3",2324]
+["3ab4497fdb166c53b35d1d864d32425996fbecfc971571646a54908b36771ca3",663]
+["b8fa4816a4036e5da5acc12db9d8af0e3912405bf48779fd3e862620e210d1a2",426]
+["9be94bdbb981243e675a7831c5f35cccb6d02e155626950ffbc2cd754be40c7b",814]
+["64a8af7b0e13804bc0538a61d97828fb4831dc59a514620e5671c506d0172ab3",184]
+["2152dc5c7e2b184659882da934d56df4e1592af92e7fa83379b988af2969b7c6",2312]
+["5f441d8fb6b1b8459dd0cd7708ff5e9c3850a2db13f388d7716e9fef5c7fd0dc",636]
+["fb9de235acdf58a18b34465fce10785a5b2e049cdb374faca0b7df35b268dd11",366]
+["03d94360097e004df0611b2d6158839f3648e263f19ab198ba45f31cb9056776",763]
+["b6a6617bcfedf0f795779db9942d084c4857465c01320eaf59a66c92eedfd1df",186]
+["635ab5ab5a6ac54eb0f3f5d57fe20433da4e7ded116f67ed5f7df2176b8323f2",2296]
+["53ee166c8e4eb9636dcb259b115f52aaafa1800366fcf3a8248369b44c824b69",626]
+["33cc29573af1a3ea63dff173578dc3ea58914c39ffba7819fcfc437038b318f9",371]
+["283cd4a3912ac771b9b10e47bce5d0a59a8e559e76bb400f4071646bffb9f7f0",776]
+["bf885bb624ee07159242bdf330fff23023ede45b898231ccca9c07c63c7b2e09",206]
+["2f20beac36c1c98d6dfb8e8d4141bf24a17ee5050d17b062f3384e67009ba04c",2308]
+["e3ae0b95742acf00018c53f7edfb9d0c83e614a214a11389eaaca4c878cc7954",654]
+["4b909ab7167c32a8c4a7483ab66759ec81aed70eb6f86756e73db1ba6515567e",437]
+["a8c3b7d71594753b4292a0593ecbd050594ccb852a64ff50e0fda016a5c7b81f",787]
+["e08c4a25ef58337472c6694d5d7d3104ac565693ef97a884aa57ebef497887f0",179]
+["ef3afb0b9c3ffa5efeeea60f7147bf68ba166cf82b3133347ba3407822a539a9",2292]
+["fa4a4298e34f2562b88d0a00713d836ea5265f3bb391c4019aaa48520006ce56",615]
+["a4abcd9e7e503fe828873c5f7e15142efa0f1c9bd8ca834f6eb0dc41e57463cc",348]
+["90ee0b8ac995c5ca3f9432ea50089476c6cdecbcde019995489141aa09fe73df",767]
+["8d601626eaebac2d4deefc332792889b80c8fa27b615b01663249e894845dcbf",199]
+["49edacc99bae0b97181c860378cceb62fb877c02a45f8dc9d5d350afed3ceea6",2304]
+["dd03b48822c989af9815fe6eb479ce417fe19a07ac3c25179869cfdf63688a1d",643]
+["5d71e0beeacf4194a46d9b15027bd7158913ea2abc0efc7bf68e994321ea6762",414]
+["b2f8b2856a538cf25a09786517604c4768f9747e09295e15c4f786af3c210ca8",778]
+["e748cb309062a3ec234bd5e41f8e6f2540207fa91db7af3ec63eb3d64927eb6b",180]
+["76c80b5d5f27ccf25c3dc90b1954a3b79514751e3d27dcb1a6e5e2ccbd90e43e",2292]
+["9711a1ef0bf4728fd3a89f0c37e47ea59e45f7efedf9e3a83a4284ec82286115",616]
+["1dc005fa31d396b9abb36449a4e8bb388b2d9f7e4a2242348b8d489e4c7711bc",354]
+["7efcb93f73642325c6d35694004ca5b364ebdbc3187b9f592e63510f71acdbc8",727]
+["f78fa07695499adc01d8e34eed5662e3528450b6122c01f1de54a31088a58bd7",174]
+["f400fb95d6f3aebc5eb7a27a4b706b0542ec955715b116734c657ad72e34ab72",2240]
+["c28d76803381e70fe7155817dd01b2171e5a1cca49600fbfc9d1da27f4c8c90a",566]
+["815d879fc35b90c56901aa9dc0f6df9d8b8b943f816ef78194df9cd5e9345da3",335]
+["5d1637f0d39450531a2f21a1f1f4d9036796993d21ce589438e0186499ea68b4",673]
+["0106a5c096c18ba32ce81820967f2a5b76bb38a7c6575c470fea10efa1d2d669",167]
+["0abcd47ada03a67813473f63aa1f2c233c79faa5806ef05bc4c49b7f58f401b5",2236]
+["5f4acbd4552d8d05a94b30a38ed2e2db2230cb1a1b38915831cc5830e753506e",555]
+["83518e1294a9720fa24f067fa9325ea0f4270b13208e6fb7248738a9710d29cf",312]
+["4623c2415399175cfa93d10096a651a3744edc80e1f75946e4cb8d4101089ec1",665]
+["b97749d8bf50048ead0a30e66d00e0fb39ab3eae19480daf14be15be585f1cd9",171]
+["e00294b92e5c8c3df3041191d1ca43f76606f7479a3f812b1c35c9db7890093c",2252]
+["184ebb5f46456e4d9932af08ca987a312f3997a33fdaf0f2d4206ffa8e69a87b",575]
+["b21b6c4d135142a83d48d1647b9951e893984d378dd7ab22490e3e3f5b00a3c0",324]
+["ebaef9fd96563dd584e080c4b3a2d9c1e817a50f4711f7316443b513139f4485",697]
+["fe6cd268a2d618c67e2a4ee794c775c381863b5697518e63e83778dfecf24b0d",191]
+["ad8220bed42d2e9fa6408cde661db4c42eecbc5bfdd8e91563a5baf86f9ed1e7",2264]
+["fe07737ea29f4dc54bf61429bf735f715e51cbdc8673e24902cae8b0912d9a57",603]
+["2a918c1f34b5153a4faf0d5512a344e9297b97ee0214df7677ed5f6b49aee646",390]
+["1c83a223e19656270bffff9055466623fde94dcf416c497619090bd5ddb84c04",706]
+["7da4d5a9629561dd78ab2a8182aff3c0e15dc5fd4859b90a534a0271e0fa3b9c",175]
+["c416727fa59db6980e7824a3bb1364cb59cb42d2dad4e798ca3cba21041ec7ff",2272]
+["0626c3aa6fb1fccc244890eb7e02d7f856c02e6e9229b81250545400571b6e95",595]
+["d8a0fe5486f8abc820b9524cbd6e776c919444678ef793b58c83316885ca5ffa",336]
+["6e66b96f1063ea1ce59b10faae42f19cfb1d7bb011ffc95efe5747cb4e75e3f2",731]
+["599f8248e3871f3a38c63000a053a4d5548918022b2e9c53e0b926a67ecde3a9",195]
+["39f7dd65f435e90a908ed5c85ce2a593cf5750bfdab4592323adcc86358c76d4",2284]
+["f85ce3ca84bf12b24628f63bc4fe0c6a2a8627c221591c7d818bd06c76c75dc1",623]
+["c1746134232bd581493a501dd0b13047a3f89a42bf3a624ce3ed7f3fea4bb2a0",402]
+["1d39de14dff8e3213041ff0d61fa5b2d9d5b173c940f34faf87cc6ff6e499da3",742]
+["4356cc03aa78d57127e3714ee690ce2ac464a8932ded21c197a508d138ff4d9a",187]
+["e88af6437d5151e7bb2c396e6d90c2f1d2a13daf65445a460b47ef42ebd892fc",2332]
+["bb54df43f358ca140f89c4b020584b798488c12535ceb7485456cb787ad801e4",655]
+["0e24340a8231b93b4d8724c70e1a1f9d16a40e3f6135736f7d23c7141638d828",372]
+["a10dce59a3b9a3b54796aa206a174714ea0cbed6eca1c494d0ed7795b4c543ea",839]
+["75c5805bfcbc6239d70899c1f50fd79fc63de865b35a30d6ba81a8a4b84a4e99",207]
+["ce7378d6c6570f481c3b5a8f6ae0a8f2d78900a87b91bfdbb5143e65ab7d1830",2344]
+["b9cf719d3c89a1695d6d548aecb4741f83f3f22fab124e6e1705869518b0dab5",683]
+["b48b43ac68ef95ae9638b3c14c16ad96f0026f47b18aa8dc38a43fa370d52eae",438]
+["4472ba1c5fd3daf4ee8995063eebd1311e1635f4b1e96f31d34de427abb0263e",850]
+["3b94f0b9b1ebaa1eb861fc3e06ea4131d827ac16e1f86c484e44aa51b1705d9d",188]
+["62e6f69da2046aa19b1e6953cc12d5d827bfd8d3d41912444cbf4a445bee9281",2332]
+["4d6dd4d48858ef37a014c357b7408edef04b1bf4bddf95df7f323c3b60ab5340",656]
+["4c7c4a2a46d8a6d35a965c6be9379acb352c0262bb588487da29828395c76832",378]
+["caa6c426854bc3d9fcd514d671c6cadbc973b92d039b98546dd68b901a06e6ff",799]
+["348ef6837868423036f10790e915a6f8c363fbd6c5e8003a443a8856286c8d9a",183]
+["7872bfe8e4ddf8a9d5aa5f86d1fd46ba019baf920645d80f91335534f887d32e",2312]
+["cf5b7a56eae23ac13c89549c274adbd3fa3224e9e0ea6bff13776d3af47af9c5",635]
+["a973f52716442322b49ca45c687d3ce014dbb625f500ee04487db06ac1ac4bf5",360]
+["9c3f71b667e5ec82b442c611d8d051bbc786d2a0a74c5bc770f5af1b4a419b28",803]
+["a8e70c57aa66d5704c49a7f32a76c73a14aebeef90ed48dddc5414cb31aa9584",203]
+["da1f89ea39a775db70c3763a6c28725c73037a09bba679e0e9a89f4c2c423cc2",2324]
+["85252cff6d0e7d877fee23706b48bbb29c7fdecf5e49494d834baf26e4270b9c",663]
+["07bcd7f8db6f42da44a16ca33816989784cf53e872f4e2584c6298e102602d4d",426]
+["feb6f8600acefbbcf9f06e4ed887f14bfb2059cf15fbe60a6f43642eb1cf3213",814]
+["9a803fc30cb3ca2c166d0ed15e69a9aa7335788322131177d87e6d7499436c09",184]
+["4c5665136c5b2b3aad061434d6e4c1d5fd5407e3a334bcb66849e01f51cdfcc9",2312]
+["6443288cb001387856a7950637578d97e37511439c24d2bd6ac693ab27861f86",636]
+["07b1e852ef0b024ee957d2e7bcc92d761802d61b24d9a2badef9015572211a9a",366]
+["97afa84ca02852ee2a2c3c6d68467a76fd697431327ded5c7f6ba4cd2b8258eb",763]
+["c8f7f93cfab74c1ce7103b00fbc557ce6107c73100c9f2a123887214785b28a5",178]
+["f06ced5f4fae66f396b6af110aa85d4129024be916e601adac191dd80fbdeab6",2260]
+["a0e9e0aa8befb08b18ce4c9e3600dca174dd3994d10f7aae2d9c19f4913526e0",586]
+["6676d6883eee961ebf8abbaf7180800c11f0d7bf3dc5127d24952e62e7daaade",347]
+["d5aed150153b0660618445115e4060663b56c041f20bbf07447cbda341e09ea7",705]
+["d9544c079d13c2ca5d888dd3a575974a096ca383b8031e3b9ea1fc8628ac1bba",198]
+["78c3e0b3fab0e2a2c7c1bff97164f45d8db2b9081f79e83ab6406f6a9047f7ee",2272]
+["cea323fd5f10531dde23f3454d536450467ae19bc39cb2297a91d174dbee8660",614]
+["84f8114fd14852b85802dbe7eeeb8612b2b6a78db6dcb98c32588178b7523506",413]
+["b8e6b957eaa7df62feedd388767b1ab8149244e5b7ae2451e8695ac037a3800c",715]
+["4bf468689b1ca6f5878f616cf7a18ac1b68ce509d0e2be04544ba65faaf9f0c3",171]
+["275ad0d695f5b41734048536d16ee5378c23cf7ef78e3de1e88e8b3179e86b63",2252]
+["dd0bca102555bcdf805c2859bf42bc87c7eafb86bd6bc035a4f3069f8e0c7b02",575]
+["1f9c270c28f713ee67413941cd1af345d1ecf324d8f5ebf396614792d9690c6c",324]
+["fdd88b2a8d559ee01ba39fc80562dc6192c20a4b8d2e11b09f5ff83eaf35f838",697]
+["33c5c652e0dec692aaee9c8277c4a9d810c0e62658611dedef446bc1b0fafc18",191]
+["9b6c87d92323a2ff34b3d226f1d31454b41c160fa10c1ec0c9d1d3ce985ed6fa",2264]
+["f0f045697d283b067f4f2df3edf342ca8e408b6fc16c73f0ea69f9e80d25ab49",603]
+["90f92379ff0c961c169b3c1708f1343c530861439654de35becb2d006a78a9b8",390]
+["cd31f0149ab9826236ede3de0066a377613e2a91c4ab9104d3a8cf3d1dc1229e",706]
+["3371293823e4248f73199538a8cf7a595d8438dc9f548aea0b211915e401d603",190]
+["189b5a5bbf731a99f20712aba35b50ce44f1c0aa34aaf48d94854c136aeabad9",2316]
+["e2ce077ecd27eb90173e8b57de281d9c76c2848084b8d46488d945d53a07fe1f",646]
+["3992440cbe78d508e79a98cedf51a478a0bf804c6735eb2ba1a0a7b0e5bad90c",383]
+["eab0cd5058727bf37f6660d11b8bad9ec0b94c90996578e25d8f8f22e41c8ce8",812]
+["3ccd53eee52663f4011f61e8dd91e0ce7cea2a8c59ce519542cc84297274f53e",210]
+["687c37096a6336d4e141035df37e7397f81757234667f15375fc98e7acd2c129",2328]
+["86be88ff90bf630a5284b9ec29764e329b91e9a8b588b1fec577288c2e33d0bb",674]
+["67011d79eba4f29f214927574b5c69843ae07d81da0758cbe836c72db0d0c9d3",449]
+["8d766a751d3a39cb0c686d21b333ecfe943d537184119f2ea5f730ecf6412686",823]
+["17d61d9e747a383213bfe1de50bd71b829f9787cc66b2e6ec29fcc30307cd653",183]
+["94924ed04165e21e526bc933f4495b732b0f183f3844a3d27216fd4101b38c5a",2312]
+["5abdf075268100561bc327b2ed42bb8ae77f3e27b84eff49117830aa994069d0",635]
+["2eac1690254d9c7726b1cad0569aa94bf1a4812bba1ec965f756d718ec1ec1dc",360]
+["d7c40b27ac0fc401b834b3e5b43f5604bc8e138fc5848547af29534a90082910",803]
+["103c3fb065a8bd4d4e4e0293a5f82aab2feef3efc2212508b5f8955c45976878",203]
+["7854faf5734738d04a5e1af65a1b26e33f33b2d8dc15444219aa4e4cbdb820e7",2324]
+["7a6f19275035793b58a164f8f295910c239174d6f52a9ff1024d6ea67e353821",663]
+["aa8bb524de29271e13941d3d7959c566cf870830b5a6745c4beac21771f844eb",426]
+["b76ac4d67ac75b961dc8e9c08e0f053e15198328701498732b48762b5efe8d81",814]
+["bddd54780add861fe28aea740de616fa7bb3abb76beb5d7ff5ec5807d41db38b",184]
+["419347fde43b89a0b957098b09a291e91bbae2049a1853eb2d1d97f2a9864ebe",2312]
+["640775ce986f57d3045e5fd8c603ffada7aefe266b87f8372b1f3c49e8287931",636]
+["7a9fe2640fabfe8eb3993dd56319e7e798d6ac15cb795a44707645814da5fb9f",366]
+["8fd0cc3f9171c02e949711c09cc0d528ae593eecd1bd44340e0363645da53164",763]
+["99353f51ee79c4cf473bf8e165e5b3d25036947b75c7240e2a7e933aa2150fb4",186]
+["2828ea6f8fbcb753e9cb435d01a105616fe0cb1d2dff3066ff0a1393900eaf63",2296]
+["4299afe1098825edce134efbe8ffd26e5093d90c34050a96cc2ab9a5997744d9",626]
+["8abb6a536192ea6f8a91a6e6ba66efdd75f2b7da5811ea7fa278d9e175909c60",371]
+["308e570885f4e712ebf5e5f8365518101b392c7c4bcc102f39d1f64cec9a4e73",776]
+["246103d728ad74569a987779cf1e4540a2a47bd456b5fdc358755d5718d3d91b",206]
+["69c104aea8d679456069271eafb29fb07ee5bb5c1e6784fdd4d3edec75c294b2",2308]
+["caed15a88fa359111fc3733d006f343dbcfa40b548e846d0c48a43750c1f82f3",654]
+["3375c06b8870962944876ab5e0b441e6728b0d395239dd02fadc582868a36911",437]
+["d88ce592ad4b6ccbfa99093417efb821104d9f34200d6546e59f7e07540782b2",787]
+["4323a725bf757d0e9e12521261017026476731052d80acdad4928012d1208946",179]
+["da75d7c1ecac040e454a1f062539f4418fe67e870803463af6f89886a6f4fa9a",2292]
+["563f1cbbffb08ec1411c6aabc0e6557338a6891090925f97c3cd1a30755f7a7a",615]
+["42ba3f504e08c568b77b90cc70fb60e230927436acb21e4e3700e5e8f706f057",348]
+["6221b92a9339a90f73894444380f4cf7b37a3768051aadf8a6e613639f493d91",767]
+["4d01fddc798cc5b8b26178d011433856379acf05361f0736d3d9727d0242e57e",199]
+["0c7e6f3fd5bcda92b48f964c8f37538c696b3f07cea8bd139b50f086964f15b3",2304]
+["a4c1bba7ca58e8c5260835a330a3ca2867e039a0dab2ce2b0aa590231befb2fb",643]
+["3d227e9390a2707adef3f50b8112a1e7fcb466dd5ca4624c9fb18dacae3ee18d",414]
+["76fad247ca59b894b420504686d1c5126c58110b3942da5b6aedebaee2988c2a",778]
+["15ef2ad9c4c801b5ae02e6d48b39ae7ba0245de94a80b38ba2420be281394cd3",180]
+["8493fecfe4b7ad84734cfc377688a21b4375f2be4cbb8134929dcad0ae7efdb3",2292]
+["3e20a16f4147237f947cc0e25c16b270e09c08345a848a91ee74d56303e09858",616]
+["66364ec40f1986120171100b2594063a5dea9b68b3cac5ca081e90f11dc2d650",354]
+["52036cf281c04bf67b7571559c622e86cf21e97c13ca8a59f2183de816bbbb0e",727]
+["d996768a9040368ca2c3a2aaa369cbba3a22b576e337782689973625aef897b0",293]
+["90ebdfa71ec0aa1102a642f2cca336fc09ad903cfccee5bd1c7f6ca69f90ba10",2304]
+["bf3343e316d20a4493ed75a1cebaca4480feff6b35261986cdd5bb28248c1241",613]
+["1a6c2389a8ffc67fb43c0843712151dcf84532be991dd72c3f82322990b9801f",246]
+["4f4c21b9cb2840e081729ff160bf8d6898c7fe7819970cae91b4548109416c1a",2348]
+["3e0a61ba0ae6a1ec7d57381560ea2720fd9668027522eca8b4bb90d59e303d3a",606]
+["7382baeca904eae20a65f6e2f60e269770f5364585d188efb63fc13f31e07689",246]
+["cb39edc6f17f96541b97f5e6ee3d3e3fff929e447ebea7a0c3d4e3e29884241e",2348]
+["e6c5ab9b0f8cf2fb43acc4833f021fe655ea98a9cc329b623fd8fffaef97797f",606]
+["cf7f3e1f5728745b0e8f245cc83b6033a501f9083c2f296f9b6c03d0ad2aa990",246]
+["adeef98fad06854d26f36d0526ffe287cd6e006ac4583d6f8f5d48f4c151214c",2348]
+["386b29125ff7c0074a2bbc1591ef00c7a651f96f703ce05e0cb19350778650f8",606]
+["0a4222e4792f6ad88fe67a99468ae881e07675d195a9f6ce59bbb88360ea05a9",246]
+["a3848cb3ecd0b4dd6da93129b04e13be353b2ceebcb3b9e8eb2827ce517ed84a",2348]
+["3ae1a1da615a9878ebea68295c48aafcef60401a3b2ad1a059f85675003e3a10",606]
+["c9a882043766ff9bf1f26ab2c5c9857ee480647cabe4c1ead68fb7d7a1d9e12b",254]
+["55f407f443473ea0e028c540422c09756dafbd7a3bf776a8fad6a83307e64609",2348]
+["79971d5e1d2b8d82a57e1cc3c70b8ec4b6f84dea42562b15ee766975dfc4b57a",614]
+["e0c1000982637972d79ee00da4d69afcfe381b1ec0f50d098bb8030e87ce0602",254]
+["48102049656b61251ca52da3ae49183c9e17c412e866ec395d38bc44575a060c",2348]
+["a0ff0b22cef1662b64a4582e1b5e89c77d10dbb5d256d0ac91ce06ac8bc29dea",614]
+["fa960cd813032aa230e35ae91d5989a622830634257c27cb8dcc88383893f2ab",254]
+["68bdc9c65fd5b1a15d4dcc5f1d83ea697cad4ea1bacd27dfa84fe73e1e710107",2348]
+["dc5b8a2930fadc67bd7a512966ec8e87189c2efec9fc7029756f6e5748b5459c",614]
+["5849eb0dfc86dcfa2b94f113cd2fe9bb157fa9037639a414c082a52f399d42e1",254]
+["f8192f06a169fc7a7eb9c289eee4be687a837003d4da6e2a7b6529ddb0089247",2348]
+["ea614f186301ee8e12fb47a1a23e1029e279bf0317aed5a1fa968c13c11a2064",614]
+["b4cf280f77c88ed5ba459f92794aeeb533da2d2a9372183621f5430f3e907be9",273]
+["333a99b073fbb6c6a0c27e75a61f95c9ad4f349f8f68e9834d992467dfabf976",2344]
+["ce3695494d5caf49160269db356a8678d3fd1dd67b924c5ba6cebfb3f7a85d72",617]
+["91b20a5340cd437c25375f7200d179c99fd4cfd7fc3af068cdcde8526d6719fb",701]
+["3bd103b641c2030766fc590dd41e46a5af2be3da37d186956a7185c5f457d691",1716]
+["8f065515fa6513b1dfda9806bad740838eefe388b7520347f8324e19efe01240",273]
+["434e97b2552047cc86f61158d80fb51cdc1cd417df6c640e9cdacf3894cf892f",2344]
+["cf1ec2853b12e2e6cd568513169a153800a570ec895219049b3ce0d7db21c54b",617]
+["d4c2f1fd50a6adcfc828551dc8182ef670b4d441c63698a23f92cd7530c5f8ea",701]
+["cfa97d33ab91be1a8f1b299f075c6d4a0cb986c7bb7ca8f3e5bc5e7e0e031342",1716]
+["7e6bd053e560b02989f5fc11458d9dbcdf0704df4e390b42322a5bab3eb66fad",273]
+["7b550e62c25c50d6a74d0b4292ff7b103dcd9895bb7d852470e3dba69fbfc8d4",2344]
+["735922c4032bfd6d27b3c7ed39ae464c8e302e75faee58dcc6a4e35d36cf26b8",617]
+["b3cac67cd8d676df7ae0c75825a8229d7aaa649864d931abd229b154f185434b",701]
+["7835bcc11bb3e27ef3c6435e3b8785fb5c4db3101d3f99f5f69ac808135820db",1716]
+["9ffeb04fa844b4cc599c4475db453848e99efb9797a4136cf9f61dbb89660309",273]
+["23b3a4fc7e06456ba0f9f8413007938865151a1b1b87df5adecb2cb126cad58c",2344]
+["2f2dc1a843263db5769b9dc44798395a67483bc421354c1d3062dc249c7b4d8a",617]
+["8acf1dcaf78e50145478d371c5b8a267fa251c91101c94aa3b41cba5dc33cbf5",701]
+["58ac9b83038e0609863bd6517165e3229864570fbb74a0195b32aa9a32903e26",1716]
+["cde4d99d5bd4a5dc3cd9fef15f17b0554ef97127ec8d9a323792abaaab1fd267",273]
+["5cef0a0ba9839a72d2ca788699b930d4089576f2aead248b2b23a48efe5da07e",2344]
+["99634e25193feb5976e892f1f5ddbb0360bdac0273b89af1a982b24e2dc1ecb9",617]
+["c2e9c378c292c9e255cadf99fca0e6e775fa59a045cf5eb9bd9ee8661ec8b06a",701]
+["86180b29e9855804d8e1ec5675b4886967c524fdc6e003022854cee4b922c659",1716]
+["6cc7d2efdad3723748e355219a80f120bdbe5500becc2bd664d1b6c56dd5e1e8",273]
+["ec534123060047b59b6e0b3dde498525feedb69b323fb2e2331f84772297b0ea",2344]
+["57cf4418a941fab6af12203cf55bd022ec76ee42178b0ca66f0e86eac44caee4",617]
+["426aefcfea366dfeb83b67a9c35c50608e2fdbea89ec93bb7c648e6675596723",701]
+["3e3150edce0578198d251a0cb03ac5bf50d4b0474d9365102cde62d17e80a6ca",1716]
+["d17e7f69fd2c558567e8d0e8bac80e6e0bb0c5e604076d311155d3183403ded8",273]
+["094131543ee5e4602ce92231ca783013bbb0a57f7930ef7019ba430af9722f17",2344]
+["8d1c95ef880a36871f11e48f53c44c49a27de09acb76ac596c7a2d8eb237fad8",617]
+["71fe0e07aea6f65b614e33a83b723a1cbc48b85cd87a38b8436909a98c2eb82e",701]
+["2b54504b58109289adfbc5d6262c263f2741ed50f9df9175f90323a87174cfba",1716]
+["1a06cc29dc196971205ef68d08214e636e49a9c57ac402e9897cc6e1586ba734",273]
+["76e55f8a9f197516242b648fc7fa4cd8c39a1ab7a98969dc3ec4f6889569e529",2344]
+["abf85fdef984717cd78c966d498f6a98a693b4e3506e526d488ab140f1e960a6",617]
+["83163762bc46c9ccee5cd58d65e95550dcaf07f5c2c035fe5a1372f0f6c2dd87",701]
+["03c892c4b9c43bed6ce68e34162bd7ef4394938e16ffc1334f4e1d8a6bff25cf",1716]
+["9986200df6c7b3eb2f50d1e49409baed07ee06485c0c337efe4294526446ae98",273]
+["c18b7fd6eee959484dcdc7ae2dfd26361be97b945039f592c6f1e988b637f120",2344]
+["8392fce734ebea20aa8de919254530d743da80642e864dcb67458134fa22c1c3",617]
+["91a19e60f9a7bc9f33c8c3a4a992872c3526446ce0ee494ab4f30cc80b6b06e7",701]
+["10981298237e0a18632bfccf86997b150c14da2b7e1100dfb5a2fda388ccabeb",1716]
+["ba9e5dd14433db165d5cc9c9acafeab957e13bbc66d8c3bef2dc26eacf9b4826",273]
+["612981a474509c6e1cbdbfd8a299ada924feb197b5e17fb34caac45fa28e7725",2344]
+["7efdb87f2982a6a7e9b5415561f6835290ea5bae4796a71433710e5174890914",617]
+["4029d9f858ac38782279ebbe6d6df9a50052d972b83d012a2e35494b35cfa0bb",701]
+["8c373c805562001f1d68868e10d84906819ceda2a4149c305f9913d451174ad9",1716]
+["fc0124daef8fbef6445ad09c863e76910b3f265501adb1c41b09a66a67b2f257",273]
+["99bc808af9120d7a76c80c0784f6d26eb5d0ec600c72cdf20f4c05f279a38668",2344]
+["22bf6e90523103a05ed4f230813bff6b8bf2a131817e9baaf716e4c20ac31a0b",617]
+["ee573b91a85ca2bb3ee12a587fc9f59c118bb54ac06dbff0341471daccb8b4f6",701]
+["672dae7aafc8f78d4ee23cd0f152797c656b5a605db4ae5e0a658b3a0c8c7872",1716]
+["1264d00911cf36708c15b90693933a0e18fd75e60d0bfca93fbd644f0f39321c",273]
+["ed81798039139922cc93a4a956b98772c6e22b38c2fbad0ed184d428bbaecd54",2344]
+["d970adacbe25fb7d8108077f2f7dec6db8733727cac09bfaeaf6ec44f6b4cd9c",617]
+["bc1eff71efdd73c97b6a1de472ff939a54a7fefc67fb03844f5f6421d02b29c2",701]
+["e955b4f04adf7776f1d077f793c406339fc213427ce206c0377ce29d40856323",1716]
+["9894d1aa941002fa58b70e314131fa205fe45e12c33927c7fa564d2a04ee12ab",273]
+["878b81ca7a841f983e41123d4b79d18861c45325c229861ee1cc2f3ca822ec93",2344]
+["d06f906d44c6896e365534afb85165c7d8c48925fcf93d24b2e595281746dd69",617]
+["ce73890bb461f40637aff5449fa2b5a7a0e5c9debf9a6d3eb8fc02888d1b72d5",701]
+["deb8c1a67d2848fd3e440423e5ff88304f1596320bd1f733eecc457718256fe6",1716]
+["4f770b4824acdf3d920680f8c30ee9f622b39d4f73e6598fc9ef22448e0b3f5a",273]
+["49e629cae0e03e5a2315dd2ef9d3cea9830546946caabee50337a847d09324bf",2344]
+["10fd6b467c9276eebaddb2a4d23d9d28a3e9a14370d91a9eefc59a2f65d9c304",617]
+["f9ef3c8cc46b228257d010e64f4754ee28a4e53632bd2aa5552b9416f8e145f6",701]
+["9c1518931d694bb2778a316223b6f18bbae78d49ff608ce348be99bc4c310789",1716]
+["7b5ddfbc305ba237785e96e717070b1050922554a47f23363976e5f312567508",273]
+["050c4cedcad798b2671685af5fdd4b728dba1e8f25a189bea3a96935c12f4a55",2344]
+["38d526d024d1f536a3095b27056f5ecea5da9bba5a0998c2f16517f43b673558",617]
+["ad977fc4fa343e49f9f5ce202cfd3ecea6dc6be0aa2ed7b684d30bc32a404c11",701]
+["596d15de5bab01f3324e65595449ba03def2acf7889018a7483110f48261f196",1716]
+["302c65524d2729403fc477f3fd1b9427920a6e4f0a66899c208e3bc65a0b8880",273]
+["9ae512ca2126b2c77ebcf51a080c353187846bd0a84c3dc4ff49771494d854d8",2344]
+["6332dea93927b98bb6f03b728f9285a0f675bfc2bae2126f930b0169a9c7d566",617]
+["df9129a7bb8f97968ff4e1ab39c33ecf72dff031692864f9a86141cf51669936",701]
+["cfc2657144e58f4d4e8224c6a155066cf11a9b0147eafedf0f2428dc7c9f456f",1716]
+["adb3ec365cf6526fc7ed4050f6b88c790cba4b0bf468c86e5f5f46b10c65e5df",273]
+["47f4f3da4e6b4a6642062c55ace6889acfd96ae0bf6d0f8eba8efe6e88a11048",2344]
+["3041cb2d418ac641fc7f1e92bc7369a36d23d3405054bd88922b89055549affe",617]
+["aba9ed8bcf4e7fb4b3a8e89a806be38a1c84a99c2b6ca7da09bb612d3dd3abc7",701]
+["7a4db379e8514a2881884f1f1cad276db9f17d48caccd1d47c9575913ef227f7",1716]
+["67da98f2403de1e2dc07ae68a89bc60f46abba153c2c6441ef8d75efbf6e5587",273]
+["ab3ae00dcd1dbf73a53c1126ed1b00f180cb08049a926b71d59b860599c629a3",2344]
+["0d8555641563f6873f23db3fe5663481b2b38461a54e5503188fbf54ad9d89b9",617]
+["30e45f3a24d28f6fccfab30349ca240c3df5da406c66c7af33bf19ee1742921c",701]
+["49a18d327a309d3d49fec02cd9ed561327b6296e04bf7a95814deedb6138d0ff",1716]
+["16c3efb6e11cff54c53dd00ee78db720bfc66d1c697cd5841f09d8acdaf80491",273]
+["dffb741b05618b167b597e248d4e68c6b1676f30a9333d728bbfacc32d76836e",2344]
+["a7675881adb5d2dcc5de3883d57e31fd1f99a75979f5a7b3e2ad5a23c2c81643",617]
+["dc8eeb25628daa681ffbe851b9d3d287fccd12632ce1f80a9a1c7074cba5b4a6",701]
+["f701ba0a550c9f460dcfc5d880b3918e155e7776e01b51928146f2593990ac63",1716]
+["3d4cc8c646a4072b7cd42e9ca1143d25d9b1ad7827f8b88011968cf0ca55a49a",273]
+["9ac8a8c73905935b04525e0a12740612effe34058e5b7eb830908e2d82222763",2344]
+["eb3cb6e299ad7d5a126860bdbeaa6837cabe6e28971dbacf8e302e49b0e13a8e",617]
+["3d6105871e991d2b69bd9eda337c69f1000820278d6ca794a8e2357a2a3e39cb",701]
+["6ce2f78ee22e2fe12ec2f6a0fe6c2c3c9e19c291f1cf3dc02059310f3de1dc5c",1716]
+["9563d20c7c0e3eba82c4e931afe2d2bc4580cc2e68a0ba9930b175aee71216ee",273]
+["0899d337f45c0a72af0a655828a626797d980e7d647cb60b1dbfea061258b62b",2344]
+["2f48f9f5aa359b38a0968ddd4e90d536886cbcba5c68e33c4a5d01690a12426b",617]
+["bcbe3d9040924d4554c651cbdc3bd73748d02cc6b9b492fdc3eb2057e3f99968",701]
+["a7bfb45d7880cd950bab0c6f621a27d99f0d39ad5793481499fb458d2220c027",1716]
+["a3d351dbc73f6504acd619858be42e862961a400cce57ea994204650fa3a60f1",273]
+["f41e1583ade6b982d1fcddb709a4123d16cfc7987be36dbf5752cb3ce958be1d",2344]
+["e57ab68f231a4bcd9f9605ce102c0c9c8947271c6860d0f86e92faa9d165ee9b",617]
+["c28429ed8a40322a856f08ce3c8b2415e9e0f318d7a9c4e0c84ae71a038606e8",701]
+["0df19cb2d16d968008756a38abdb13c031f5bd00ed37c3372ede0a3886406e7b",1716]
+["0f3ef46f974a0d87c6f6bbc3efde54454094e233628fd0c4594161aaae78d4e0",273]
+["568b55e1b61715f9175b90477697b3210d0c30923969e5cb58cac04f1f32d628",2344]
+["7c4f67e69e9079c9c00dade967171ba8da086f594ab3556e3af82d5c3912878f",617]
+["4a8b4eeaad9344e40f325017f58b3ace82fc1ec4602f22ab4c99be78afaae14e",701]
+["b6d2f99176f9cba2952a785f61437e7f681ffff7c692b5918496f01c37d36e9a",1716]
+["3e365cf3d11596bd61e177a83c1b08c71ca8e4a694f579af920b35ecf71198a2",273]
+["56fed2d75a4410c2b980d37b7fb4f316e547184a990852f7d5c9a295770af178",2344]
+["26a28108a02b2ae5657a716d58ce8c7189d2d9c7e27a6dd6dff52e5ebdfb0514",617]
+["a7a12ca30999b58bee8c4d09ab4db288f1729447bf4f3bdee86e8d1278602da0",701]
+["df8acfa029c7ba33c5e5154bdc0d5d41062063a30c09638a72a5584f71094485",1716]
+["991eae431eeee168928f7127f6b30eb49aae052f8d5e90b8acf40b7eb0296589",273]
+["3e9245a8861d28e25c76d2f207456d71cf11d68b871cb640aeb7774bf1d03044",2344]
+["baea81735d59c26229adbbe59c3d9ad45a9778541ae1cd344da2bdf449721cc4",617]
+["fb0b25345f2f5e82b09e76caa7696b5bc37a510fbd75581639f7b4e15f429e36",701]
+["87dd8bf520ac49fce5f974fb36c02d3d17c3ace093485e680311082440e27f96",1716]
+["bb10e6bd6c509bf3f8189ec0d6fe39837771fc9178ca6fc11e79e3a3a7526ba0",281]
+["d32497c0b9efcbfed87036687c6480c596411af1efdc374749e3dfe76cb92ace",2344]
+["208f3295b1d91abc7a57fd8048b86960349fc2e88e3acd8be076cddcf4d35eaf",625]
+["6c3527b2d299b661b5f298821478306820f31835a24d7a18f8fedb164616ee13",753]
+["7683975b0a0ced522a71b7d7d991624f17b005a6f077f541b555c7899b164362",2054]
+["d253f304306fa0c286be0ac629888708271ac63ab8e133f834b189be823d425d",281]
+["1ede33ebe05bc38ad81bfbbb4d905d834d36fbeb2ff4d4070451f7ce05dce059",2344]
+["4696b678cf6f57ed309654fa92ef03f47231aa7a9c1aa38232e2ce34bec2a751",625]
+["246df3ee5153756109f20154c24d2f366c248b095e4ce8cfb654a41eee0797b2",753]
+["308b9bebb7a82a6b15da7a28dbc5f90b2b4aef8d2dbf31d4f41a209a86978817",2054]
+["28ed9a935e1bb6695a61e4e672e13a3f412294b7b9655e3c2112c6ba88dd34a6",281]
+["528af3acdfb57921907881eea6d9f36e07f46da0a8ff1ef8c84da6e6a9084ca9",2344]
+["e05aaece5d75ba005dc90de49ce1962123373d53f6dff479b69d481dd07b2fa1",625]
+["34933df1a3dd00f2cdcd9de944d94fe1ed8959b9c11a9f1e175ec8f1146b75bc",753]
+["0d2003eccdf6c42a4c81424db3898f323caa2b51e5df917001e8627f9b93a757",2054]
+["0d25870d4e08d1e54236a951b268ccd8f6facfb7b74de16cbc6453ff4e890656",281]
+["cb8827c50402c7d86ccac811e48571c58b3841cd4cca5d406bbae3a35c5dfae0",2344]
+["08e991aecf2cccd104deaf5fd025ef94dce929bb91c6afaac5741b11ce01292f",625]
+["d2687d7e59583b8883a2375789c1622968aafbca92704fade8686fd527d68cc1",753]
+["a5dcb80aa3ab585a071a9371df2270056d3006987205c8cd7993268059bfec69",2054]
+["b744aa534fc5cd7d05b8795d339d17c1b72128824cd96175a783318f7b78cd9a",281]
+["beecbdde3f50fe6547dcb061194fb4163639177150750bfb5489d959f86db732",2344]
+["5e7a1ed28c7e18b18712c77d23aca4b3b78d834ee7928a72947dca5e2c0cd505",625]
+["90b9f508dd3febd24c9faa4cb1ae9d3c1cde31a313c2ca7f4325c4d9f12b7280",753]
+["cc6b3350618c0b806bd752e8ad494b2df2ab863011784fe38b1f8c2e611d7d00",2054]
+["5ff1f02f529074c434241cc6eec1531e47c5f358a8135e55f2e6448579599ce4",281]
+["bbe11a2b35074830b2f0ed702fe23f999da2c90813671de84fc9cfec3d235d3d",2344]
+["ea378ff7ada12b10aab5e80cec8c4c794e62397632a596d503f6b2a260f29442",625]
+["bd1a2f7d3716b126c740ab660fc68c6e4709304fe23e68f666721b4ffa8e79b4",753]
+["10a502b36f598949c1ce7b90a44e9820c9432941731ee68231aa74ba0c0b0d2a",2054]
+["d07b6bcb2f4a4004c5ed7d91b874b5582f383358c2611bd2cbea3c5f09481b64",281]
+["aee39873f99cf05c11ff15146ec0169930f620ad4f11230fe070eace0e0ef778",2344]
+["072b6672a7af83a10b617f4e345fac855f05c145c514026d87dbaed900567d36",625]
+["4c02816b6b72cd33460258edd090c48f68c1eb5b9a84b04bec23185dde4426ed",753]
+["5f4e9d04359aac81fb085d654beebe4bc8ddf1c724125a331724ad90b1332a96",2054]
+["83622a2b2dd3cdeba88249bb557e4e4f27bd6ae6a1a7d754e0960f6198b5b111",281]
+["8a94d7fb32a1676d509dc4e46bac2520750be936392005c878541d1e2dd55b00",2344]
+["61ae09def0276c5e889180ad480d0863739a5d50943538d47d48a3736b45c0b0",625]
+["987409d02c6f1d3c4f3129602d9c1b04b38b0deaefad17771a7cc750752bd66a",753]
+["941cf85fa7caf28330db1acb4ad5bfb332e9baf78c8bc0f8ae11cf9dec75d940",2054]
+["870480bcad0fd11b2bbfcabca543d9b3cbbd761ff35176a3a455650a4d651533",281]
+["c3a8981a70ed8fbd7ba85d043ad56cbf2646f8d2292fc4ff85d51b19eeda7d5f",2344]
+["d98302d2d383d2232179c575665dad68fd1e4376066395132fb75ff62a369da5",625]
+["9b94175734eb003e61b745ac3531ac7094e78110ec92202bc3477eae7e0a9673",753]
+["d1e736c7b42973b66c3879ee491f1b2972350e02f8b8dbe5c28476719cce723b",2054]
+["e750ece2e10d09de4a00a63264cfaf923f773f6ec32fd6c0385ba572f2d59806",281]
+["141a688a6ebe671131a89cbe77fccaf48150d8511cc3f494c618905a8e97c48e",2344]
+["672dc6dfa83a7e973906c4ecdc3a8f56ed0ce2207967b4fd986955c8f24bab50",625]
+["d2a32a7224dbc4190eddc94d74b0bc3c6838576ede4fe3b675b81f375c9bdbb7",753]
+["0b6a584740c910a90808e724c8d3630007f118e4af388a7bd93582108748ffbf",2054]
+["d7c01a11e422d80f4928d0030cbac48cda4e7f6636ae24455f773eb3f25ec388",281]
+["de1a8d4fec8fd51ec7c3a2550859fd0271f7286a56eacb55c704e45b6db158bb",2344]
+["97e3a7f42697106144512e1e4788c84cc21c9141265c1d37aaa3928eb710d652",625]
+["ebc5b4951c2eebc1d1e3ee412c988d0cefc412616194e6d0e641e675bf9abf21",753]
+["7b454e66ced209f98b384e1a1956fc9818c59b609143b896a4986df3cba9fd5f",2054]
+["48b947de5753e3eb81f3f1e9fc36b6fa50189d3acbbcfe3254fb81581ab98f2e",281]
+["7133eb185f38611455bed2de495cb678acdd43bf1b0c7ae2293fa7ce01946b0b",2344]
+["9e554010199164657d1494d5b589b813bc484144b1a1df91349dac82eec64cca",625]
+["2f9a05f56af32a2abf85bd5fb95ef4f1c5e2940f3132d4fc534e839bf394ade6",753]
+["413f26ecb01964e9a4179a61295f9ba1f9d35df2e840dbd5fc53611335ae0205",2054]
+["65b0d67dd7d49eb0e8a9f3df15e5150b8d66972b3a85489a02a8b54fa1675dc6",281]
+["2a2bf85f9c66ee35a5cb52a4f6507307d0b14b5d28e2720762c8601e002eb77e",2344]
+["363b7d579893f52e1caff5af2de2bd8e058ddeb3b669d6bf38144fb0c656108c",625]
+["ce3d500adfdee05c31e2ab7bf0a21c3045803e5cf4b19814e3bda92f9b94e356",753]
+["3631eb0a7ca7fe27d976ac5c54d5362f05f5f1cdf2ef02683ffbf6854157cd12",2054]
+["bbc7035de82a28c5dd2aeab34bb9d6b8b397479a2a70da63372277bd948c68d4",281]
+["d8c9f8e014e3452d814ac68a092faff6353779df9e8e6d6e7dd785f68ac71c0c",2344]
+["1904ca8122393ac3f90c5d618676fcfa29bb1418009b965cad8b7fce06ced18d",625]
+["4986a669222966dd2f0d04998d1326fb9869b21a5e17cd080d7f98ad3eee4609",753]
+["17610504c2518d644f713c552bf8691d675338d23e0011dbd5365c44dd171f64",2054]
+["7385da6b31865cbc59780544cf8a14f9da16fc73032fa505620c37af9545a57b",281]
+["be3d7411597703e7ca381591e63456ac4a2a973e3d88312c27adf5d2799542a4",2344]
+["7e712d7969e11cdacd896849402b137239b30e9bf8fe74c783867633c901d3aa",625]
+["11e6cdcb9a0c378ddb31755bc458c82cad2cce65fe2e4c8b61a811b24c595455",753]
+["107b9b13b1f1381b49934ac613192b0debfae2b222c3f946c8264dbe096b220e",2054]
+["386c5ff0d21ff9519f78a3955938f225854bf3afd6dfa9c668806f1f159952b9",281]
+["03cba3e06af7d1ca12b40f518cc499908e030fe93251f8ae30372f1a62eac567",2344]
+["27d5d6178ace1189a201cfc745829fd87b3d9103fe92ea1a0ae993f1587709a1",625]
+["d7c8ebd2703503b92a4ade32b69b2fb986aa94eafaae0d3697aa381c72d1d34a",753]
+["4f0669a19f631e39f685752359a6f152f73a23b3c7f043a48631abab084e3285",2054]
+["cdb3e8ce33c3eb670d13e281e09c2b000da7a9dde912a8ce3fd8e6eb8bf99757",281]
+["7c6ffcb60075a2ecabc3cfb7f4a42135debfe67651f9db7000718759ead3e55a",2344]
+["30b9b37e3795a3300d773a156937e4d37d877e86db56fe407b556658e659a0cd",625]
+["20d274641949fb1364a538f34fee6b2b0045e6aaf9c131e07d655fe17d975552",753]
+["dcd12970a554a4ac5553f58b711807b8aee50a08322fd69d2d39d22a666b881e",2054]
+["e499a932d342aa44e4167e2c34b577a702e3af4b8a655e88b07f80710f249495",281]
+["7041b80e7dc0cc92a3a0ff6655f69bb34cc3a5bce19cf4ad3edab7ab76d176ca",2344]
+["f331aa97deb69f23a37597529910e88cc677e51744ee760b7db7f59becf83fb2",625]
+["bd247fc253617f800a496f8cfff8937064fac8c0f648e02c376c6ba979ee9761",753]
+["180fbe48be1453a423409492844a1bab10bf4ba92e7eedd6bcc82ccd9344e495",2054]
+["56a3e15229e92790855fdc4249b0c24da679a849d64afa6ed96b9cdeb50e203f",281]
+["c4e029cebc5d161ee3a5167762c434cf07f5dc53849a3879fb903c7a5850b975",2344]
+["c7372e2db7a08688e16d53b1e6b0828430647c7d8471d5998104ef46fff48570",625]
+["be160c5ca471a97aa4a6479c7e474541ffde20c6dec28594c2454a11be72b89b",753]
+["da78a9c0c04ac5f2e438d2059888260d291101ab0e04d4bcc6346c142e49d0cf",2054]
+["8980346fc6701fc609ff7f694f342fd5b6dca5f66f83e2bfc14de990f9e70e35",281]
+["8685c5ddb7a389ce8cc525f72abb780eed290d122c4813fe81624d2b339cd905",2344]
+["927439e75c4a64df3946a6a21a6b9832c81cb823ede4b5471ecef3446b2989ce",625]
+["fe73779875774635f10f928873a2174fd0f44d32d0beab7e84eff76ffcfb7401",753]
+["75087af341575add1179c3a2905c8e91f2e8554448f662d350f435e2cabe6e4e",2054]
+["55b39282a4d5d85cba7c0667cede61170f1394f36fd24e3c8a20571cf0559313",281]
+["4dfcc1f4eeaee16a9fbc06b56dcb842221f50a6437277b48b7bba0ad05abd017",2344]
+["916f6555319d7c245a6cbe38fb326b4351eaf890b7e94ce691fccd98cd0ff3f1",625]
+["75c314ebd1a667bef89bdf37db6ef514fd7d11c64f8fe550a73cf9bcf7af2979",753]
+["269bd3593a7d1c640e7e0a2682d6ef420652c3ea066573b1ff962a7f266eaab0",2054]
+["f5f63d2ade2066485980844e216866fc6544b058cc0bddcc5338bd130a9c308b",281]
+["40d6883d581967ade38315c0a73861fff2b369d52836e2878bf33117ac2b96d8",2344]
+["ab8ff3c40c1a829394cbb3779ec6c7753312983c01f5f7864cf145abe7e4ac3f",625]
+["6828a8b9426e4e8a9b2752015ea4416dd0622b6c53f8f85f94d5796508e6ba83",753]
+["d8325c15f921c0c4ed246c663e7ab5dcd603de4cd0b1d5ab3afc284e87eabf44",2054]
+["f5ed23c41eefcb12a193de2e7537debcc6c5e8d528a44a1b60fdbd275eceeed6",281]
+["532ec4d030a9224d828584e1f9b0d9878e2adab9cd6db2aedba4a044e431c62e",2344]
+["bfb798bfb81b333456e4afa0e1002a47a9e5bb6261389ec57892f132a86d2d78",625]
+["54699009902fd44a51fbf80620ee9f501e3d9af988632d129e5a2547c7f32f56",753]
+["9907f503f80b359c14244031139b42a3d2c8d92b7fbace62b8488adf6228eb8e",2054]
+["0a4408414e4b1d7bfc101c273b206e627e5cf527e72f207112cfe42d132ec11b",281]
+["7d715250cec6f3abb7b99482c606262d006e4fc9d970cb2e43c9e266cfe36fa2",2344]
+["8edf8b39f1702b059ee7f929d5a32e1abf322339de3cab88a2487bb11634c1ec",625]
+["b4855bdabd692fc6368c280aa54860248817b8705e4bc9b1a932aa1b60cf8330",753]
+["7fd4658e7f01d0789cf5d4c64478cf6d2d80e8f0db60833626dc7f5ccddf184e",2054]
+["5a1845c22ebb9bdffb4e64e41bce94890ecb6a0152a498099f34e011a642defb",281]
+["90622a7b4e916938feb34f4b779b94c83e733eea36c65e61268bb4d2a2f46e61",2344]
+["5f52a1aa1da36d325747a4796d15aa628d7804c0f00d194ca6368cdd8c91c1d7",625]
+["6ee6b10c3b622c9b74f546704a0ffc746248612b647601ee094d6eab838c8502",753]
+["846f0a21a0fa54caa35177e43d34bc34c11a2daa508fe0de446b0c5c605ee7de",2054]
+["4999fd6d961b61a1a5bbc6ff9ede32dcc7a32d9b08d0179b4903a35688b79b45",261]
+["d0e1b1cb910e3e76c3fc0b751e3a9700c9f44045059b02ef8d1d24923234fcd3",2348]
+["1491f8b35b62681e94c00293ad567cb821495c7e1b3c77e690b61efbc88d2ce9",621]
+["82ae7d074d90425acd519d06d6f1f640379bbba165f65f1b978e77eacef55cea",261]
+["87a73f48f9481fd03c30bcfa6fc61af1b212e26530ab18af716b9734d6b72df8",2348]
+["4879d6fa01c0fd3593c799b9150839fca62be5f390519d0cec3f36ae7b4c22bf",621]
+["58f4b37edf1c9c8eb2eb6b9f51eacff7c726191593c08a4486d5c4e337378085",261]
+["25a7877734d79ecc2905e3882140d14ce50250138a8136c5830ac236f3cd1127",2348]
+["9b4efb94bfa2746ea606d2fb69c377d296ce1737bdfd7cbe472bd5e73cda6365",621]
+["5e23e4d001dadaaad1cb7526f5f8c448cfc8f6405442a0c8c5baf0dcc7a385e8",261]
+["1fa12aa849ae6c90bc7e946f516c4238fa5824fc9f4e7c1a6dad2a0d48d26587",2348]
+["09701f4bc7c399b6637296d5c9a18fb71787a1418909ae54d3c4d3c3becf1498",621]
+["1f7c66199222ba14bb5492e2d83d3b4761de01e91281a389351126accf8c82de",269]
+["d92a5b86e36bcbd4a317a8075d53d3d0dd210aa35a1c8de877ef2b26cb256c21",2348]
+["280d4c4d76e76e080d938887f79d5efbf3681daf48e0101b8f0bfba6603b9d24",629]
+["238bd5097aec3b7a2a11527d533c351d152493ee46e043f2de4224d5fbc35e40",269]
+["bcf44e97f1bf741725f284d66c72c0c30bab04429885a15b6644541ebb156392",2348]
+["fb9d5eed543983cba6559de9798ce09ec41eedc829438f3293cb9d2cb9c316cb",629]
+["dd3f6100c034177621170d601c186ea904e1ff52862703bcfd1dc7845af969c1",269]
+["426a7ca0e2f9c1691ca84e63c8ae5d85510799e5963691685b5d512c1d3c89f5",2348]
+["aff38f57dd1ceb91831e58a26942598c11c23827dbe4006286601524adda209e",629]
+["7f7fdf9b88ebbdf9a406d6acedd3f0042a5d14d33aea6f89577fc816a7cafab8",269]
+["449673483276eeb525a815942dee263ece9b49e50d6b4c48d5dda993d32647c1",2348]
+["11766761d152d3203fa503b293a2fd7834a7b4b54dabe3ac9d5b4fe75b5dadcb",629]
+["b7641ee9a0d05f2bbec2ef8f0d0750ed9598057049147b797c926bf43379cb3c",240]
+["0729b87491ec04e2eeb1ba1323e0e453f1154768efcc7df4924f412b2383c864",2300]
+["5d02da41fa83ca6d0fa2d0852448a6bb6ef0ea8f2a82756b983a0f0901d8854b",560]
+["c83100b44b298106c33109dc4df2d98a7e9bc6e213ec7d1dc92c810c7803e7b2",256]
+["1cbc179738fa88f3d984685e0e4493aa28c1ee80176aa10b4578b19236a77237",2292]
+["26c0188313416406c94a05ac77da4bde9864b3b34bf55f5cae657c37498fb99a",572]
+["afeb1b611b97798377272731277cd1e7b360309a44c91c45c2c4619963e6b20a",240]
+["48f5ee238acbf5fd779d7736e66c9d246b4aa05b3ffa5b25338eb8c2eed5d3af",2300]
+["bc9f85fe4e43635d32a5aa6be989643eac3e4ebe11db5e4384cb42aeeacfc423",560]
+["a69083b49db9d1976cd7d4cc2a2f9b38da9ad6dbd7d2eee1f14f01a6e8d6619f",264]
+["7e1af59150df732192f87f70cd965f7a93e40ff3d5021d48d0e2cae079f54430",2296]
+["1c7133629301512d9b3ca7c7a4c8e93358344aa33d966082ddffc43e4797d050",580]
+["bfce0ff99040af6db8722b067038ea80c0a8cba06c92d2843d5b59b69c30bfed",313]
+["78aaa40f030456cebf74d12d6103d0a31a6de2b38326a493ffed1fa2c17a9a97",2516]
+["fd671e9f81abc446580b71bb7490588328944fa5c1acd89c9479a75f9f5c29c3",824]
+["15b9279afa9771b011adcdeaf7c661cf427272cac5a5410767aabf375d608919",313]
+["a98768e328b4d684805366d8427e59e6a0ae0f27898d30db49cb3f8789235b0d",2516]
+["04abd1c1a2f28d0fbda729271da29d85a9d1aa2c9b7f6d6d7ad2db1329dccdcd",824]
+["8402fb00b31870d92b3bd00ec61f637eca0aaa2dd48c3117dbe459d9e45921af",313]
+["d012626a16b0e55ff5c2a51f540f8cf8a220a5f001fcbc1180502645d2b82c46",2516]
+["92b8550271611c739ebf0d8183101b6bca74f9ee94a9a1ea38dab2e0c563be63",824]
+["19b73e02e2ed72c8500a7af61c0a3bbb527a23254b5ae1ab01f28794220eba78",313]
+["6af301525b2dfb619e6dee3c14522b564211f5460cc3ee0a5054fdb00eda771d",2516]
+["fbd1dec9fea6f3ee823e00d6427f765e00ecdb2298038f96fd452f4c6fda2153",824]
+["5c5f24afca9bfd46a07dcefd43e27b5bb1020d7e39c39827be7b0cd29a173d39",313]
+["33fd9fbc8d4eb58c0500e8e4e2a21256b1c1e1a661330085edb9a76b2eec4b8e",2516]
+["2864a5208dc33cc1fa61c30b1073617ea64e758ebf88b6cf3a50e063302046b1",824]
+["5a5d129c043006c83f8a222051d78a4ee8efd4e91bc6552f9e70cd06596e93e7",313]
+["1e9b4fa8cc519e23489d7041e06b4af1c2e20ec6bce06237800f0227d09d2cb7",2516]
+["e59245d436257210e3d9227c140b14120bb039f5487fecc0ca3aa2ea48606906",824]
+["e762ad34aa8c2914235e32e1c01101533c973726cc521f6862863b1a3cb3fb3a",313]
+["91b319269598b7348de06afaf09907d0df05418bcd51d4310e1276df0ba9cf9d",2516]
+["e15b53566ac42d7d0b1d1fd479e930d61bb17e074473f3db3c9a0ebd3a465a92",824]
+["0ed1f37afdc99159ab7a0d2a3d9b4badc38aa828084f83369c491dbc67306127",313]
+["04c1df5fe262e7fb91f1c64fb909c8f5c9cd111b1d0dab18010d64723ecfc788",2516]
+["2dd527ed27fb97bc63d039f8715f445f67df99dcb476c2b296e74569a085fa31",824]
+["f791ca63726aa45682205dff1cedfb82a021c8cac8db4b933ffd1ea8184f9cc5",313]
+["913f105c62f6a41de39de2d009bf946133eafc6762330055a80443bb9540f804",2516]
+["eb5cc027a2b3a8c7f7b9e845b40d28d0ef5f8a1106b245cb64849de83e4b84ae",824]
+["7501b45b768c40bb5920df969a4c2059cb6b80321cf52c91cd39c08e3690e04d",313]
+["2253dddf21f9db5434298b9815ea5f9c87ccf596cceec4fb564e0f68c9210ee8",2516]
+["ecba0b34d8bd4d0c8401ac75b27e540d07e5ad708b0ad931b2aedf56959d4477",824]
+["0e03a7ea1f5c4d4753075887ee135d365078e3c4550b8dae80afad2f38881854",313]
+["893b5c1703cbd7f9fcc0d919eaee370a4c56db10f302c859ae1020bdecf0c79c",2516]
+["99303f3fbc8146f66d64c28c10128ad919f6c3f12b6442a09c87c663f1247fa9",824]
+["de89b5c9586d925d820479707d864ef2a601951dccebedbc71866bf83464c1b1",313]
+["e3cb70e6c0dd21b316dda64fd3c602f50d6df4da5652b1b831e8f953c5cb8da4",2516]
+["dfab492c33eaa5836eccea879ed229102b2cd4b59cfc27a8f23d1b2b248fa653",824]
+["cf82b17680b56790c1afbbb7f1180062c3e41c33bf9d7b94fb8d92629ebf990c",313]
+["b59bebefc28d727d927e7d868bc9e48093eb44e6dcf8489dce5c50a1ba7df827",2516]
+["6edf4054eff9cd2020a9a7b1e169e94120669bd4be94e075e9d9fa06aa1ec7d2",824]
+["08be33afe4c210e8109807b4b4590b729512c5ada6aa267be749f68333ccc061",313]
+["fef0b42a91041cff7b4818cff395efd1bb18d9dc70283d8f9d980046883921c4",2516]
+["d4a74e72e9e0589859e354849ce2fa999865aaac034496f81d8b08d4f94e3d7e",824]
+["26ba27cdd1253dc67093d876e73581613b292026ff9d7c446ac788ad24d03948",313]
+["8189b91b3860bfdcd5e490d9cc5d193fd14cdb5e90a001d1d578fd8b3aba3013",2516]
+["2023dcd9c158549fcfacdb214f0ad19b36766721b0b42e0b223de43d0c05f994",824]
+["5b902a3eda60807510600c628243c8293d986aa5f46a14756cfa0d6191d7e8b3",313]
+["b21f6be3c41c14358532e4339255db88321c4cef5bd511c8be89d24ca2c3eac1",2516]
+["c5f655cf077cf22dc89e0fe8e18a63538722996fb07a05100c39166ad21ca399",824]
+["6e1802e141740eb91cc26aa3fd5fea007e2d3de199818061be67e6ac5dbfb19d",313]
+["b5b7d9b88a1e6bf82bc523006684a91abe2eaa449f514c5956ef0e32dbbc81d7",2516]
+["f88b7ed6a47bb2fa91e2fc1759e4e4bc20092791dd15ca1b8e6920bc8a38ee32",824]
+["fe5c456cf0eb05f42190c5934c960607d181b04905109df8efacf31e783b7ed1",313]
+["508bffb8071d5b762e31bcb2764abb1548973f28e4910d0aa402308141e566bb",2516]
+["65c6f5371a9fd1a6b167481da25fdbd2827cdbda5c533a824b2bd3f3ef47dd4e",824]
+["575de018272d19de86877768a5b396cfa968dfae723d4d94f75511ebff91fbe1",313]
+["ceb9a626eabbe6b399e9723a5bfc1135174442735e42de75fe4889bd1b25c940",2516]
+["f4cf21a414e90902e4ee46af867335f64e0c3098036345cddf87ad53a967c21b",824]
+["71939090da6257ee0fd6b70610d546b2d20c834cf6ff839923aa0ad8de40dceb",313]
+["3fbae2e57903f6b40f01271fd8ba4341bc992df0cb5d146c858036cd94e8d013",2516]
+["bfbed99d5eaf7faf23f61e4dd14caf74577984c429f0535b5ae1f7b143d67f26",824]
+["fbc2b536b117593f14498e99f6541154d2c0d6cbae99a08b089302bc8ba2707c",313]
+["46e5a02ca880e41c5afdfd18e1af5b432ea82e2942cb5b56b45174efe06aa832",2516]
+["808ee8a6e832919fb92747b38e0b971c438e47fb9ec391444f571945060514fa",824]
+["753ddf4c848e377d74ed835f6e5e9f571007986885761b86ca1c18de34b655ab",313]
+["064dcde61335be86b65d48fa6a2aa9846bb284f1b8d830809e72c457a2492a9f",2516]
+["fd1167d6db86fc8582e06c770f01a896c35c0084e2e233f2a961f0d6638c2aa8",824]
+["3b22335fcab97c32a93951109966409b8f4c9ba868c55214159a49160d9b62f0",313]
+["7af2c3bdfd3e0902f223f953e2ed4c3b95ec0ffb535256c39bb50b679cf7dbcb",2516]
+["db38870deb9ec7ae5691f55bc741a064a848c984ac99368a19e3bbec892fdbd0",824]
+["be6fd15a5f196398c14fcbfc2ebdfdbe5ddfd28ce9f66a40cfb645e55ebd39e0",313]
+["2ca477c7ecc7dd91097799c78195ced49fa8814670923c6f4e2856d46a977935",2516]
+["dd7a7ba7f2f73d36cf6f77bf24035227854e48d5351b619f77130c9c68fd3691",824]
+["e4ed0b3ae0cf16a1a16657c3a40dbf66b4d327a203c4febcc2bd76b365421e3d",313]
+["ae43261ba2754dd79cf7e137ca46a357a37a21cb27787ff3807ba6dec9d202bb",2516]
+["071f4588350cf6f4f99bddeeb9164e5fbcc9c1680acec76a3fb268c4fb1d2927",824]
+["356d3730a578f356ac375ddf592b03e462b1d86bb9f07af766064c9080a5bb43",169]
+["e5dbdbd6cc7522e29515dd80bb4976b71d69d46306fd38719eeb28edd5283205",2268]
+["0fda61406ce90585e77ae57ca220dea3bf56159dd4b55e3ecb04f8de2709dae5",567]
+["a09a38c82903b1b6e624b0d204e738b2270c9ad391178f97ec20637cf4e50c0d",171]
+["e84ed0ec8b980948b6c93b528141fa33776a7a6e55f537af758a13bd4e0f51e4",2268]
+["368be0affd238dc508905eb1f020cad165e501f2af35f5ef6236d31cb9f77955",569]
+["950e52f2632c0dd439887337b26a42f887601ec66382a4d969bad07373966f5d",173]
+["3e8959fa24594121f7412d37fab5bdeecd7dcb4ed54ee5b543a9dcf245b6e197",2288]
+["9e94ddce331da0c4de6d16b5c384f9f754d1b3e9bad18f99d69d409e1b2d37fb",587]
+["2a9be68efae5e30037e9c341a7d512c4f82caf6a27e6bdc747c41f003171c11a",167]
+["2209ee3309af159d61a42a757228a954bd73b6783aea335e0ed6b89008e0b295",2264]
+["7678510fe59d328eb2d404931addfe561d2777b6dbdd9a952845a11393044909",557]
+["20fbf12b87e1a8758735996b06962985418acb07bafe2400681decdc0d8e2007",171]
+["7917f48dec938172f2c3fcf680c03f0a85d08049f2631a03939e2047a19a32c9",2280]
+["261f83617260dd25c11ed4612bad9869e7674d85800be0b63df0a518426721f5",577]
+["362bf3b64f37bd9314139249603b33833eeb51721dea42317a25c77dbc03f10f",623]
+["f34f7ff2aa85f89e6bf5d06034bf7f6f9ed9a8d69c600c28f55f531d5bdffe94",852]
+["f6157d75c61739579cded52d225e97e77589b7a820f9d873b2b36cde79400514",623]
+["9df9ffe99d1e2ddf88f8a85cac0eb718eb50dedaea30130fc8c703e2c5c35ec5",852]
+["07cf0a68c31b1f3a0282134d0e198cce58dfe9869d1f468ca56e6e01d52bad25",633]
+["14794a6acd689da74561337a0db5f244f9be71c0eeaed0ea9fd9a66bab2bbf9f",897]
+["232dfe4317383c9ccddb7fe6972f7c1bb254470e39a4407f36001a6fc9d45fe0",633]
+["266e883b4b968763e236ec06a5b6dec60f3c03c722ad5d8bbe13ffb2cf670cbc",897]
diff --git a/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/captures.jsonl b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/captures.jsonl
new file mode 100644
index 0000000000..aaca7f8839
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/captures.jsonl
@@ -0,0 +1,2076 @@
+["abs_bf16x2",0,0,"sm_80","+ptx70","sm_80",[0,null,null,1,2],"b240082a3d30fbc94fc932aab7e42ddc344adba7bbde5cb146ae6ef75ce6fd0f",0]
+["abs_f16x2",0,0,"sm_70","+ptx65",null,[3,null,null,4,null],"60d898e51204ebe3671593ba80714848dbc581e88736f2bda5a90885c8a9acc2",0]
+["abs_f16x2",0,0,"sm_75","+ptx65","sm_75",[5,null,null,6,7],"caef89546c514abbd2a762ed1175810f3a288c0e3056b639146fb349e1fbb340",0]
+["active_mask",0,0,"sm_30","+ptx62",null,[8,9,10,11,null],"d8295717c32cade0c82dad409c666b6a47b25f2f710383e86bcc2718a65eccc4",0]
+["add_bf16x2",0,0,"sm_90","+ptx78","sm_90",[12,null,null,13,14],"dc2e0fe40e6ea1919d3c41faf36b4ec56aa3990e96278137eec0856424ff273b",0]
+["add_f16x2",0,0,"sm_70","+ptx60",null,[15,null,null,16,null],"5e3cf1c2534e8e4a1ee9fc7c9cbeefd43b743b230e7ac47f0cb1c0e3ff3181db",0]
+["add_f16x2",1,0,"sm_70","+ptx60","sm_75",[17,null,null,16,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["add_f32x2",0,0,"sm_100","+ptx86","sm_100",[19,null,null,20,21],"3cc02dabf66441d4e56cc6b71ab5661f1bfeb2cc79aeb9eb5516992d98769d35",0]
+["add_ftz_f32x2",0,0,"sm_100","+ptx86","sm_100",[22,null,null,23,21],"5976373db6845ee5cf51ecd0a1dc21e6dd93fd18b15f880f8584195e59f305bc",0]
+["add_rm_f32",1,0,"sm_80","+ptx70","sm_80",[24,27,25,26,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rm_f64",1,0,"sm_80","+ptx70","sm_80",[28,31,29,30,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rm_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[32,35,33,34,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rm_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[36,null,null,37,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rm_sat_f32",1,0,"sm_80","+ptx70","sm_80",[38,null,null,39,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rn_f32",1,0,"sm_80","+ptx70","sm_80",[40,43,41,42,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rn_f64",1,0,"sm_80","+ptx70","sm_80",[44,47,45,46,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rn_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[48,51,49,50,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rn_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[52,null,null,53,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rn_sat_f32",1,0,"sm_80","+ptx70","sm_80",[54,null,null,55,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rp_f32",1,0,"sm_80","+ptx70","sm_80",[56,59,57,58,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rp_f64",1,0,"sm_80","+ptx70","sm_80",[60,63,61,62,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rp_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[64,67,65,66,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rp_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[68,null,null,69,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rp_sat_f32",1,0,"sm_80","+ptx70","sm_80",[70,null,null,71,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rz_f32",1,0,"sm_80","+ptx70","sm_80",[72,75,73,74,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rz_f64",1,0,"sm_80","+ptx70","sm_80",[76,79,77,78,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rz_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[80,83,81,82,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rz_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[84,null,null,85,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["add_rz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[86,null,null,87,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["all_sync",0,0,"sm_30","+ptx60",null,[88,89,90,91,null],"2f0ddf17f0ab8fbff9dc1943ee906125689b12817bfd9d24837d83c2c5a9d385",0]
+["any_sync",0,0,"sm_30","+ptx60",null,[92,93,94,95,null],"a7cbac25c02c0a2153ed12cfe55d70ed1bd840889231b36a70727619432156e5",0]
+["ballot_sync",0,0,"sm_30","+ptx60",null,[96,97,98,99,null],"4eb9bbfe06eaa716fe362bba924e810a2a3c0cfeb85df6e4ac30adf1928b2bad",0]
+["barrier_cluster_arrive",0,0,"sm_90","+ptx78","sm_90",[100,101,102,103,14],"620fd48878d75451af54efebe32a3d096587dadd54bca9a64b7e7bd9936693e6",0]
+["barrier_cluster_arrive",0,0,"sm_90","+ptx80",null,[104,101,102,105,null],"a2b319303c60b00864fac1ec3433943aa92526c830bf27b9938e20f77dd5cc3d",0]
+["barrier_cluster_arrive_aligned",0,0,"sm_90","+ptx78","sm_90",[106,107,108,109,14],"d4b7b86084a2793f04c22f7bba900e09bfcd6aa8c1cc398174f8c144d2c7e68f",0]
+["barrier_cluster_arrive_aligned",0,0,"sm_90","+ptx80",null,[110,107,108,111,null],"a0ca80e1bf59b2ad44e7bb3800e413504a2fc6fb8acbedc08264a6a4d471e9f6",0]
+["barrier_cluster_arrive_relaxed",0,0,"sm_90","+ptx80","sm_90",[112,113,114,115,14],"1b709d15eb9d24f41d7058b4c3111a4a26dfdbeabc7b5da5799eea3b5f953324",0]
+["barrier_cluster_arrive_relaxed_aligned",0,0,"sm_90","+ptx80","sm_90",[116,117,118,119,14],"26cefe6cfdf83db80a92d3aaa0bb73562602c23cc2b616bff519b466c1fc5da1",0]
+["barrier_cluster_wait",0,0,"sm_90","+ptx78","sm_90",[120,121,122,123,14],"4a7b70144141dfa1815c69e343d35e3176ef01ebb3304f643db721b06694f2d7",0]
+["barrier_cluster_wait",0,0,"sm_90","+ptx80",null,[124,121,122,125,null],"2947feb7f7e14cd688881c2c19213fa01f696338ed6f05c195e1b9226f31d06e",0]
+["barrier_cluster_wait_aligned",0,0,"sm_90","+ptx78","sm_90",[126,127,128,129,14],"229b7d8145a3286de9e952e49d1ea7ddce53c133d11a02202d8bb23ef8ba98e3",0]
+["barrier_cluster_wait_aligned",0,0,"sm_90","+ptx80",null,[130,127,128,131,null],"ebdf35740fb489df053c9cc9034d9bdd4ec4806a6b493b6d8bc945c6630ce9a2",0]
+["barrier_cta_arrive",0,0,"sm_30","+ptx60",null,[132,133,134,135,null],"5153539d16bde8d03ed086668ac8579239657935d5f1b6fc3673b7eb240b32a0",0]
+["barrier_cta_arrive",1,0,"sm_30","+ptx60","sm_75",[136,139,137,135,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["barrier_cta_arrive_aligned",0,0,"sm_20","+ptx32",null,[140,141,142,143,null],"20a2d37da8ae5f329cf72a4ac87ce312321351f303788e08ffe2b562c812ce84",0]
+["barrier_cta_arrive_aligned",1,0,"sm_20","+ptx32","sm_75",[144,147,145,143,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["barrier_cta_sync",0,0,"sm_30","+ptx60",null,[148,149,150,151,null],"aabb7823805c13b501d92c287400265ce505fc325d192af22474d4376c7b6cba",0]
+["barrier_cta_sync",1,0,"sm_30","+ptx60","sm_75",[152,154,153,151,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["barrier_cta_sync_aligned",0,0,"sm_20","+ptx32",null,[155,156,157,158,null],"6ab69d10243a093fd075f7be5dc9d3d9dca3cd0fc7582105c152f3e2f155ee8c",0]
+["barrier_cta_sync_aligned",1,0,"sm_20","+ptx32","sm_75",[159,161,160,158,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["block_dim_x",0,0,"sm_70","+ptx60",null,[162,163,164,165,null],"a047944cec8f711074a5b649fc885b253cbc42eb5e886163506038e6bd96a14e",0]
+["block_dim_y",0,0,"sm_70","+ptx60",null,[166,167,168,169,null],"7a743b8b771e6b4416bec54a4bb48f1bba5b7b584cabea833e006dee21c90976",0]
+["block_dim_z",0,0,"sm_70","+ptx60",null,[170,171,172,173,null],"809d7dca22de97a82a855513b3d711c5affba352692637c74b27f3d1e25fd2e5",0]
+["block_idx_x",0,0,"sm_70","+ptx60",null,[174,175,176,177,null],"fcf5cd1fd3e5fd6e906634dfbb6c03739d24a629703f743bb6c3869956ad7612",0]
+["block_idx_y",0,0,"sm_70","+ptx60",null,[178,179,180,181,null],"b390522b0252ec1e7085417d1e4f6b1856b6c5a1b8e78d547b575ebbaa3e6055",0]
+["block_idx_z",0,0,"sm_70","+ptx60",null,[182,183,184,185,null],"168a5545b150c1fe93d9c94845e5d2dbbec16800aaf78e8ca4fd7828d425c071",0]
+["breakpoint",0,0,"sm_20","+ptx32",null,[186,null,null,187,null],"d6250c4b4efb0b8a4970a9f4bfb9c519aab6ba5274a6abdb1871b97e2cf587db",0]
+["breakpoint",1,0,"sm_20","+ptx32","sm_75",[188,null,null,187,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["clc_query_get_first_ctaid_x",0,0,"sm_100","+ptx86","sm_100",[189,190,191,192,21],"32f888a8a6200a2862467629f22c19976c90aeb28c8762d6422b81676faba514",0]
+["clc_query_get_first_ctaid_y",0,0,"sm_100","+ptx86","sm_100",[193,194,195,196,21],"05b1e3b9b6eb2b7b19895651199ae9f35c4fefdd43f99eef25db1791cc8b922a",0]
+["clc_query_get_first_ctaid_z",0,0,"sm_100","+ptx86","sm_100",[197,198,199,200,21],"a686d0f8c84905ba266546f64212e8730504a6d2987db258439949f3c6c36d1e",0]
+["clc_query_is_canceled",0,0,"sm_100","+ptx86","sm_100",[201,202,203,204,21],"c73a807fcb7b9c79d0054b85c63496142d58cd7d4b25f3ec53e849a23254085a",0]
+["clc_try_cancel",0,0,"sm_100","+ptx86","sm_100",[205,206,207,208,21],"501524d6faba61cd6501bef8c4bfacdd337a5fc7d70cbd79b8039eeb4e529eb3",0]
+["clc_try_cancel_multicast",0,0,"sm_100a","+ptx86","sm_100a",[209,210,211,212,213],"7177864a34d996298c01b1de3a36d9906a4c764fb315e335c7117214ce23ae7e",0]
+["clock",1,0,"sm_20","+ptx32",null,[214,217,215,216,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["clock",1,0,"sm_20","+ptx32","sm_75",[214,217,215,216,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["clock64",1,0,"sm_20","+ptx32",null,[218,221,219,220,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["clock64",1,0,"sm_20","+ptx32","sm_75",[218,221,219,220,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cluster_block_count",0,0,"sm_90","+ptx78","sm_90",[222,223,224,225,14],"8eb2f45d10de8cfb72a206e9c33447383ec0a2e8e8585202a92d3772ee72b436",0]
+["cluster_block_idx_x",0,0,"sm_90","+ptx78","sm_90",[226,227,228,229,14],"3ec5d51cb63b41c38282ef148e4eb57935d7b1639e118a5579279951f526b7b8",0]
+["cluster_block_idx_y",0,0,"sm_90","+ptx78","sm_90",[230,231,232,233,14],"aa645f69a42ba809a6ff57078e1f2e55c7189c1f3880bee0a08b016acd40cc88",0]
+["cluster_block_idx_z",0,0,"sm_90","+ptx78","sm_90",[234,235,236,237,14],"b926aa27ac13f2a4143da04ec08830c4019a3f24018e4dffc84f737757a962bb",0]
+["cluster_block_rank",0,0,"sm_90","+ptx78","sm_90",[238,239,240,241,14],"291823e93e8a189fb5020a6a15d3f53dc0a06efcf69c3f025b483a9e022b8f1f",0]
+["cluster_dim_x",0,0,"sm_90","+ptx78","sm_90",[242,243,244,245,14],"6f361ec1f939661addb54a978094f0f0c3b3e064066598e438c6534693b08963",0]
+["cluster_dim_y",0,0,"sm_90","+ptx78","sm_90",[246,247,248,249,14],"a7277cb46f93f8afa5818eb379d178c8c06bcff0f9a936158d9ec84ed3e192e5",0]
+["cluster_dim_z",0,0,"sm_90","+ptx78","sm_90",[250,251,252,253,14],"232cfccbe04a5518c9fe452161f85a58eac38b61544347f955a1d256c0859b22",0]
+["cluster_grid_dim_x",0,0,"sm_90","+ptx78","sm_90",[254,255,256,257,14],"186f94a225c452f1a0d4236898c07fe835251022c3b891e85a631a1f5ac811ac",0]
+["cluster_grid_dim_y",0,0,"sm_90","+ptx78","sm_90",[258,259,260,261,14],"c5611ef7a8c38cfec8e2238fa8d7c92df5785cbbbe6b7efe09a6990f1e0ddeb2",0]
+["cluster_grid_dim_z",0,0,"sm_90","+ptx78","sm_90",[262,263,264,265,14],"ef39123e61e0674b0642c7795de40413eaadf47bd1fd9e735773fbccb3edd67a",0]
+["cluster_idx_x",0,0,"sm_90","+ptx78","sm_90",[266,267,268,269,14],"6379775896f02a542947f4da6b7a875b89579cb4e99cd394ecd2a1abcbfd4b8d",0]
+["cluster_idx_y",0,0,"sm_90","+ptx78","sm_90",[270,271,272,273,14],"a50c708daaa5c70cb82bb7ca1cd522afc21ddcf810b05ad887a9ea81677c5db2",0]
+["cluster_idx_z",0,0,"sm_90","+ptx78","sm_90",[274,275,276,277,14],"c9eadfce2c364c5bc151a823c0f0ae6c62af558d885dd8fe9d0bc2dc7d251f64",0]
+["cos_approx_f32",1,0,"sm_80","+ptx70","sm_80",[278,null,null,279,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cos_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[280,null,null,281,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cp_async_bulk_commit_group",0,0,"sm_90","+ptx80","sm_90",[282,283,284,285,14],"3a55fd146981cf8554c5242d95c26f1c09b2d4a36681923a345f64a2ee8c652b",0]
+["cp_async_bulk_prefetch_tensor_1d_l2",0,0,"sm_90","+ptx80","sm_90",[286,287,288,289,290],"e56c95dadca9f34fb539ea7c03cb591252f43ce7dc2f08c6865b93facdec096f",0]
+["cp_async_bulk_prefetch_tensor_1d_l2_cache_hint",0,0,"sm_90","+ptx80","sm_90",[291,292,293,294,290],"ccde4baa574b70c2c1d5745756b76a7279a32038961e53ae6871025b84f01156",0]
+["cp_async_bulk_prefetch_tensor_2d_l2",0,0,"sm_90","+ptx80","sm_90",[295,296,297,298,290],"49a9c7883b44a043b47472798b0fbc860dcde8661742853fdac3603bd60e8bbb",0]
+["cp_async_bulk_prefetch_tensor_2d_l2_cache_hint",0,0,"sm_90","+ptx80","sm_90",[299,300,301,302,290],"82b405bd137d72b339e112cc4cc3a790699a174c57a826b44a20041477ceb945",0]
+["cp_async_bulk_prefetch_tensor_3d_l2",0,0,"sm_90","+ptx80","sm_90",[303,304,305,306,290],"35534400642663b61d524ad09d7df4d386b95d1ff72113bb77eb9f739f003ff2",0]
+["cp_async_bulk_prefetch_tensor_3d_l2_cache_hint",0,0,"sm_90","+ptx80","sm_90",[307,308,309,310,290],"74f92dd6927a3cc07a51a084ce91f46a39a688331ff2d438285dede1fd3b72d7",0]
+["cp_async_bulk_prefetch_tensor_4d_l2",0,0,"sm_90","+ptx80","sm_90",[311,312,313,314,290],"4d93f26db17edaaf65b8a7f4d8d3e85af15abcb572cdf215d15b4ca976c8ac3a",0]
+["cp_async_bulk_prefetch_tensor_4d_l2_cache_hint",0,0,"sm_90","+ptx80","sm_90",[315,316,317,318,290],"bcfe101bf27bb2a85726f85de59a12594d3a43b5dafd6c47a92212ddbd93c1c2",0]
+["cp_async_bulk_prefetch_tensor_5d_l2",0,0,"sm_90","+ptx80","sm_90",[319,320,321,322,290],"4ba0fb3484d4818d7b942655b713e0325996d82917f5ef39de8cde4a94198633",0]
+["cp_async_bulk_prefetch_tensor_5d_l2_cache_hint",0,0,"sm_90","+ptx80","sm_90",[323,324,325,326,290],"6d9abb13e97a71dc1873a0a5e5f650f89b37e5485dd90d0125388eddb5805476",0]
+["cp_async_bulk_prefetch_tensor_gather4_2d_l2",0,0,"sm_100a","+ptx86","sm_100a",[327,328,329,330,331],"8e8b400848fbef42d3fe4cac7c9e8decdfe041d63c8f2f1243584aeebebb8e89",0]
+["cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint",0,0,"sm_100a","+ptx86","sm_100a",[332,333,334,335,331],"ca80ef54996837341adcc9afb0ea7a2347e8459d2769fbf98ae41f80342d6b34",0]
+["cp_async_bulk_tensor_1d_g2s",0,0,"sm_90","+ptx80","sm_90",[336,337,338,339,290],"5399c0f968ca55c3dc37bb6244a623f2a2db8a15f5b5302b40ba80cce8954ab3",0]
+["cp_async_bulk_tensor_1d_s2g",0,0,"sm_90","+ptx80","sm_90",[340,341,342,343,290],"52366bf8b0109fde7975367a3eec15d40b9ccb3709d68f516f2d6741c9744c5d",0]
+["cp_async_bulk_tensor_2d_g2s",0,0,"sm_90","+ptx80","sm_90",[344,345,346,347,290],"6e5fee868e38648404b1c0fb2859b48fb60f88b9c880799c957d6841e80b525a",0]
+["cp_async_bulk_tensor_2d_g2s_multicast",0,0,"sm_90","+ptx80","sm_90",[348,349,350,351,290],"bf3e596ea98195a16b4a49bc26061977593f763454eff4daf6c1ef2d6f971d4c",0]
+["cp_async_bulk_tensor_2d_g2s_multicast_cg2",0,0,"sm_100a","+ptx86","sm_100a",[352,353,354,355,331],"36a7ec9e98967ba93c54105eff4d097af7ea3909d9f5c5d73c5065484d88ab02",0]
+["cp_async_bulk_tensor_2d_s2g",0,0,"sm_90","+ptx80","sm_90",[356,357,358,359,290],"a8e93abe577c031d9546000ff0e62e0a65e64da657f3d10a6610f878b74bac94",0]
+["cp_async_bulk_tensor_3d_g2s",0,0,"sm_90","+ptx80","sm_90",[360,361,362,363,290],"c0abcace69bbe468b15b4e4b04a3c1757573ba873895787525a694f07d408842",0]
+["cp_async_bulk_tensor_3d_s2g",0,0,"sm_90","+ptx80","sm_90",[364,365,366,367,290],"56be562f498b26d7d2299503ea5f8cd454f221d8b3d36a949b2d327149e90806",0]
+["cp_async_bulk_tensor_4d_g2s",0,0,"sm_90","+ptx80","sm_90",[368,369,370,371,290],"580927c420cd9a69baa59aea5b9ac611978b16e730bf655f1411863e62efba19",0]
+["cp_async_bulk_tensor_4d_s2g",0,0,"sm_90","+ptx80","sm_90",[372,373,374,375,290],"84b3b55d6584e5c5c6daccbecfdd93d464a23d0816b561adf86dc981d26a9856",0]
+["cp_async_bulk_tensor_5d_g2s",0,0,"sm_90","+ptx80","sm_90",[376,377,378,379,290],"3420dce2fb1a1372978b1444a67e5a50657e617abfdff1abd1bbdf5585c440d2",0]
+["cp_async_bulk_tensor_5d_s2g",0,0,"sm_90","+ptx80","sm_90",[380,381,382,383,290],"c7c8576b3a4e15e1a7cf03361d00e8f4c3d89fe8d3d3da30473856a237f6b291",0]
+["cp_async_bulk_tensor_reduce_add_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[384,385,386,387,290],"375b93818be5f35fab04fc1e3aa46907489dd4c1dc1ebbd9bda04560b69c5dea",0]
+["cp_async_bulk_tensor_reduce_add_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[388,389,390,391,290],"b5e4c4499d95a0f384fad133fb5104da10897f1a1cfa2af84f3ace22400effe9",0]
+["cp_async_bulk_tensor_reduce_add_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[392,393,394,395,290],"99c9b5a0028327e068d2008af4db17434ebde9746afea63ab499ca319814389d",0]
+["cp_async_bulk_tensor_reduce_add_tile_1d",0,0,"sm_90","+ptx80","sm_90",[396,397,398,399,290],"c288996c89bde61b387c8995187b54124b23996813c080e60e3bfd8ae83f95ad",0]
+["cp_async_bulk_tensor_reduce_add_tile_2d",0,0,"sm_90","+ptx80","sm_90",[400,401,402,403,290],"ab75bda2b99a3d8b71e0eebff6d6285505ad227d1b4cc6f501e7eb297dbbbf32",0]
+["cp_async_bulk_tensor_reduce_add_tile_3d",0,0,"sm_90","+ptx80","sm_90",[404,405,406,407,290],"d16d0f6e6ce4ff621acc0a82d30a0c94c459a3d9e2921533823d4f6d5d238674",0]
+["cp_async_bulk_tensor_reduce_add_tile_4d",0,0,"sm_90","+ptx80","sm_90",[408,409,410,411,290],"9113092d092e6d14a61379d0eda7cc5e277c62c0df7dc3cd20c6607fc4dc2690",0]
+["cp_async_bulk_tensor_reduce_add_tile_5d",0,0,"sm_90","+ptx80","sm_90",[412,413,414,415,290],"c1c8776c78d91c279ad6396ca3666d909a3e000800cf68eeaae5598f2149e7c2",0]
+["cp_async_bulk_tensor_reduce_and_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[416,417,418,419,290],"528041a4bf8bc657c50b1afdf5f3d3b524361d7439f8f5d98e06fd592ed191c5",0]
+["cp_async_bulk_tensor_reduce_and_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[420,421,422,423,290],"270148648fdbd9ec61f6f29b60ff2a0adf3fa0427e0743c7c33d541f5b7aadb1",0]
+["cp_async_bulk_tensor_reduce_and_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[424,425,426,427,290],"8fd76cd86368a4e6fd11a8b3ccc7b2dc805ae705763976c5741aa0af956e9f62",0]
+["cp_async_bulk_tensor_reduce_and_tile_1d",0,0,"sm_90","+ptx80","sm_90",[428,429,430,431,290],"6ba9cde9f89432ca3b12ea863db60646eaf90337ea54425324fc0c9a54d584cd",0]
+["cp_async_bulk_tensor_reduce_and_tile_2d",0,0,"sm_90","+ptx80","sm_90",[432,433,434,435,290],"c8d82e750f596ef4ec1081e6a9b40465e72ba1a5f6b0f27ad4e6b22e35a95049",0]
+["cp_async_bulk_tensor_reduce_and_tile_3d",0,0,"sm_90","+ptx80","sm_90",[436,437,438,439,290],"61f95ea60a6adabac5fc07eaa3fcef6d9255e737c78be21927967bb3d2acd9a4",0]
+["cp_async_bulk_tensor_reduce_and_tile_4d",0,0,"sm_90","+ptx80","sm_90",[440,441,442,443,290],"ee7829041d50e82479b86831af4e029785bbf00dbc84975b936bc529971b9525",0]
+["cp_async_bulk_tensor_reduce_and_tile_5d",0,0,"sm_90","+ptx80","sm_90",[444,445,446,447,290],"b35c40452803658197de869909b1b95511329752704915a86c74e582e76e0823",0]
+["cp_async_bulk_tensor_reduce_dec_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[448,449,450,451,290],"839a918353cb28de73534291d11aef34a855c8edf6c2a433324d47eb5757960f",0]
+["cp_async_bulk_tensor_reduce_dec_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[452,453,454,455,290],"ade80d4ef342247059798ffd61e5dd631177e54b3cc46556d4a7b7fe0daf37a7",0]
+["cp_async_bulk_tensor_reduce_dec_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[456,457,458,459,290],"68b8b5e137f0ef57fd568218fad3305ba0374f56b8a5349ae57d57a9f7c770b6",0]
+["cp_async_bulk_tensor_reduce_dec_tile_1d",0,0,"sm_90","+ptx80","sm_90",[460,461,462,463,290],"e41669f87a8f5b44452b16d6ad4371d0ffbe8117911da79095a84793a220595e",0]
+["cp_async_bulk_tensor_reduce_dec_tile_2d",0,0,"sm_90","+ptx80","sm_90",[464,465,466,467,290],"18471ea1cc16b264fd61adfc2af9b1041a38f159c17760ab6fde6f5560b76cf4",0]
+["cp_async_bulk_tensor_reduce_dec_tile_3d",0,0,"sm_90","+ptx80","sm_90",[468,469,470,471,290],"655593efeb7dccc7c5b1a854216fcb74c1659dd24aa7a76d63da9b212f32ee75",0]
+["cp_async_bulk_tensor_reduce_dec_tile_4d",0,0,"sm_90","+ptx80","sm_90",[472,473,474,475,290],"a6e436cdab071b14d5113f2fd6fe4e2b02c5cad10cd5b2d350017df2f56f80af",0]
+["cp_async_bulk_tensor_reduce_dec_tile_5d",0,0,"sm_90","+ptx80","sm_90",[476,477,478,479,290],"ac0b8da8f9c2a3394a5861c8ab031a1d973729f126eb052050a3c29886cc7944",0]
+["cp_async_bulk_tensor_reduce_inc_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[480,481,482,483,290],"b32f1b12d224f5054624637b8c84b21b16828e78df068f0b91fec7cd268c1051",0]
+["cp_async_bulk_tensor_reduce_inc_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[484,485,486,487,290],"d035240ba9ddcc1ad96ec3ccdc8334b5e211a8cd42673990ad25fd3b69e99e3e",0]
+["cp_async_bulk_tensor_reduce_inc_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[488,489,490,491,290],"b724b082cf06daadac0d72de41f0313c2277a406b244ad8f36bb9dbd358eea3f",0]
+["cp_async_bulk_tensor_reduce_inc_tile_1d",0,0,"sm_90","+ptx80","sm_90",[492,493,494,495,290],"88b78969c3fe78716d9a7125bcc586dbd23be0bf62a4ff405098e87246395a4a",0]
+["cp_async_bulk_tensor_reduce_inc_tile_2d",0,0,"sm_90","+ptx80","sm_90",[496,497,498,499,290],"be692463af9460a9ac7c0f6abbc68db5f4f65346d37de394b348a470605344e4",0]
+["cp_async_bulk_tensor_reduce_inc_tile_3d",0,0,"sm_90","+ptx80","sm_90",[500,501,502,503,290],"77d782783877683ac280f4c63cb3a9fdad4ee8f33f92f03ef51555998ec867f5",0]
+["cp_async_bulk_tensor_reduce_inc_tile_4d",0,0,"sm_90","+ptx80","sm_90",[504,505,506,507,290],"bfe447654e03cbac7a87965ca359ce0623c68d67f9b7cb5b5c97c6ccc41bfaf7",0]
+["cp_async_bulk_tensor_reduce_inc_tile_5d",0,0,"sm_90","+ptx80","sm_90",[508,509,510,511,290],"8dc584464329be941d2023cb4f6d12bf3cfb2aa3d601766dcb01e13bcb1b6713",0]
+["cp_async_bulk_tensor_reduce_max_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[512,513,514,515,290],"297421a5a06a572ed424ec0290f6c531b9191e543a7c4302e00a78437fb50143",0]
+["cp_async_bulk_tensor_reduce_max_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[516,517,518,519,290],"b77618b0a95dd91046dda10f5133fb3fb7aadbf365a6a53ba7f18b280556a7d3",0]
+["cp_async_bulk_tensor_reduce_max_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[520,521,522,523,290],"798c8fa5ab006c9c4cb4f6c52cb95e465ca0e8c2d64f5251dd3919304841b52e",0]
+["cp_async_bulk_tensor_reduce_max_tile_1d",0,0,"sm_90","+ptx80","sm_90",[524,525,526,527,290],"a9b0106411d0336398ec3e87c25360891e0d77e084c82f437d8ccb7f219b68a8",0]
+["cp_async_bulk_tensor_reduce_max_tile_2d",0,0,"sm_90","+ptx80","sm_90",[528,529,530,531,290],"6963e78361ca92904bbe6f8af1a09b670d384ec89ddea78abf8f64ab0c40dcfe",0]
+["cp_async_bulk_tensor_reduce_max_tile_3d",0,0,"sm_90","+ptx80","sm_90",[532,533,534,535,290],"3391714665e42d83e5feb25893287945f14fc7b15d56ad83d9ac023c2ac86bf9",0]
+["cp_async_bulk_tensor_reduce_max_tile_4d",0,0,"sm_90","+ptx80","sm_90",[536,537,538,539,290],"7b7505e933868f75992176ea7348e8333b894b908572e81434abc2b0949c17a2",0]
+["cp_async_bulk_tensor_reduce_max_tile_5d",0,0,"sm_90","+ptx80","sm_90",[540,541,542,543,290],"6886eabef9e335086a66e18cad520c8e667880898746cccb4e0b20b3d2e11a5d",0]
+["cp_async_bulk_tensor_reduce_min_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[544,545,546,547,290],"33f33d4f94cbd2ecc4c8b146a91e7ba6b0d15ef1eb28840ddf7c4288012ab0cc",0]
+["cp_async_bulk_tensor_reduce_min_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[548,549,550,551,290],"5752be0acd86f5a1747487081070529ae84aee8c7f6c548b5412023b80f33fb7",0]
+["cp_async_bulk_tensor_reduce_min_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[552,553,554,555,290],"39b312aba614c65ad6a3da05cb95295298824a20b259f86dd98ad08de35b11fe",0]
+["cp_async_bulk_tensor_reduce_min_tile_1d",0,0,"sm_90","+ptx80","sm_90",[556,557,558,559,290],"ada2c97a694faa6f696696cc7db26b87c590e8c78260ee9a6996de44fb8b2ebd",0]
+["cp_async_bulk_tensor_reduce_min_tile_2d",0,0,"sm_90","+ptx80","sm_90",[560,561,562,563,290],"bcd54195d21c71b23ddab15186c8eef156a9d2f40b80b147545e1df7d6f9269c",0]
+["cp_async_bulk_tensor_reduce_min_tile_3d",0,0,"sm_90","+ptx80","sm_90",[564,565,566,567,290],"123c885489f0ea6f655c6a2a9b47f4245d79f477e0fd8b067c83613a8f9661e7",0]
+["cp_async_bulk_tensor_reduce_min_tile_4d",0,0,"sm_90","+ptx80","sm_90",[568,569,570,571,290],"e88dbfd40bfae4bfee200a466926835afa884d6457dad0bed77499c32a180993",0]
+["cp_async_bulk_tensor_reduce_min_tile_5d",0,0,"sm_90","+ptx80","sm_90",[572,573,574,575,290],"3fc89c072949f145c26c811fc3485a47c8b55a526cb64dfda0e005da20165412",0]
+["cp_async_bulk_tensor_reduce_or_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[576,577,578,579,290],"8442b900bddd492d181077477766d2e763d53445e4d6fc1dcd3ffa5360b17610",0]
+["cp_async_bulk_tensor_reduce_or_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[580,581,582,583,290],"53ae688570e7eba2efdecc5f3396bbc12ca7a34fc6afa8b5a064f9342ef1d08d",0]
+["cp_async_bulk_tensor_reduce_or_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[584,585,586,587,290],"dfc4815492e31337d79ae0998b572899ca8b4c74f76dea83b86d843fab9250c2",0]
+["cp_async_bulk_tensor_reduce_or_tile_1d",0,0,"sm_90","+ptx80","sm_90",[588,589,590,591,290],"0c946f7a18a9db836018335269d75efed174a1acb211a98d479c15fac83cc573",0]
+["cp_async_bulk_tensor_reduce_or_tile_2d",0,0,"sm_90","+ptx80","sm_90",[592,593,594,595,290],"3d9bad7c50333c2ab89f2bc1a8cc23374e503fc7930d87fddabf0d122b75cce1",0]
+["cp_async_bulk_tensor_reduce_or_tile_3d",0,0,"sm_90","+ptx80","sm_90",[596,597,598,599,290],"4572b4c373a631c12bf5799381a0f29f86211a5331b3ad6a725f9517e4de47ac",0]
+["cp_async_bulk_tensor_reduce_or_tile_4d",0,0,"sm_90","+ptx80","sm_90",[600,601,602,603,290],"746a0ab741472acce2f60d9f86fec253c0680385940129d36eedec52c9c6a4cc",0]
+["cp_async_bulk_tensor_reduce_or_tile_5d",0,0,"sm_90","+ptx80","sm_90",[604,605,606,607,290],"3c2598c9e44dbc3c515607842069f4b337cd3820e008830a43040856010cf5e4",0]
+["cp_async_bulk_tensor_reduce_xor_im2col_3d",0,0,"sm_90","+ptx80","sm_90",[608,609,610,611,290],"15e1631daf0c4b8eddf51ea9a3de73e90a089e4edb1d3d89c9ec442efb28beb9",0]
+["cp_async_bulk_tensor_reduce_xor_im2col_4d",0,0,"sm_90","+ptx80","sm_90",[612,613,614,615,290],"1ec6dc5104b6c5a5993d1e09af9873de7f5aaab83e6d471d561f0e4e8924bd09",0]
+["cp_async_bulk_tensor_reduce_xor_im2col_5d",0,0,"sm_90","+ptx80","sm_90",[616,617,618,619,290],"fb41f6d16dd56e97b8f379a27e76789835a6efd5cc53ee54cb242736814f52e9",0]
+["cp_async_bulk_tensor_reduce_xor_tile_1d",0,0,"sm_90","+ptx80","sm_90",[620,621,622,623,290],"0df4727a9d091bd86888789f158574ca50c89d4c8c166179b98167e47438a354",0]
+["cp_async_bulk_tensor_reduce_xor_tile_2d",0,0,"sm_90","+ptx80","sm_90",[624,625,626,627,290],"8bc288506ef00c91d77830de568492e635c2f9e2a4afe037cf91ad66d378c85e",0]
+["cp_async_bulk_tensor_reduce_xor_tile_3d",0,0,"sm_90","+ptx80","sm_90",[628,629,630,631,290],"efc28e00e37a634782a726d0d6bea4c305841973415f41ea851f16b9c101cb1b",0]
+["cp_async_bulk_tensor_reduce_xor_tile_4d",0,0,"sm_90","+ptx80","sm_90",[632,633,634,635,290],"0385ba7b96c3f5dfade2d09dceade00163d0f3c782033846722f91dc011e62c9",0]
+["cp_async_bulk_tensor_reduce_xor_tile_5d",0,0,"sm_90","+ptx80","sm_90",[636,637,638,639,290],"9d01c321bf8966808d53907fa0460cb59069f9aa2528b42e8e9af97dc7c167c4",0]
+["cp_async_bulk_wait_group",0,0,"sm_90","+ptx80","sm_90",[640,641,642,643,14],"d1be2e9973f5634e2007ee92744d034baec69173db760ececfbc8cdcd78ff85e",0]
+["cp_async_bulk_wait_group_read",0,0,"sm_90","+ptx80","sm_90",[644,645,646,647,14],"6aab2df2343322d39297721fd62b0b884babdd03769dd5778df6edae39c5aae5",0]
+["cp_async_ca_16",0,0,"sm_80","+ptx70","sm_80",[648,649,650,651,2],"e34b77ef6e46919df30e854314e31baa21edaf6250054c23187ec1bd78aadeda",0]
+["cp_async_ca_4",0,0,"sm_80","+ptx70","sm_80",[652,653,654,655,2],"8af07954c7c818fa6e59fb72f9d42c3fdf13a92b1aa760c3469ae9e87033d118",0]
+["cp_async_ca_8",0,0,"sm_80","+ptx70","sm_80",[656,657,658,659,2],"8d5f99368a77505dd5a1eaea8f06fd43fc2d7d1ede04f024dd9de18abdb605ad",0]
+["cp_async_ca_zfill_16",0,0,"sm_80","+ptx70","sm_80",[660,661,662,663,2],"4f550b590fee54e2128949281d844e0e86f472dd6e1c78a1a8a4421925b88069",0]
+["cp_async_ca_zfill_4",0,0,"sm_80","+ptx70","sm_80",[664,665,666,667,2],"55a36301c9cba24a7d1771c433d646e6b973260ad42e84d5b75f6a2bc37aea94",0]
+["cp_async_ca_zfill_8",0,0,"sm_80","+ptx70","sm_80",[668,669,670,671,2],"ad20d7b006e8e6bec4e0e23f9eee6c5410812c9fe3713438bd959461db9b6110",0]
+["cp_async_cg_16",0,0,"sm_80","+ptx70","sm_80",[672,673,674,675,2],"9bdccebc44b15ef79fd934cfd1533a12b441c736230955a99b12b85bede2eab4",0]
+["cp_async_cg_zfill_16",0,0,"sm_80","+ptx70","sm_80",[676,677,678,679,2],"91825fd222feb25f7e220cd093ed2fabdf870d948854caee3226dce6c0d6d0ab",0]
+["cp_async_commit_group",0,0,"sm_80","+ptx70","sm_80",[680,681,682,683,2],"eafe49e3ca0bdd06f683f448b18d73867f64f3d402be0b5f972aa5179ef93d63",0]
+["cp_async_mbarrier_arrive",0,0,"sm_80","+ptx70","sm_80",[684,685,686,687,2],"bd323e71d020757409372ec3d6e0dc6fb74b8b502265cf699bb97ec64c89f5bc",0]
+["cp_async_mbarrier_arrive_noinc",0,0,"sm_80","+ptx70","sm_80",[688,689,690,691,2],"3c0ac4f3f2a7719f39c3e6d873572b5dd0eec66c764521e13b63c1311eef97a9",0]
+["cp_async_mbarrier_arrive_noinc_shared",0,0,"sm_80","+ptx70","sm_80",[692,693,694,695,2],"4f20bc6fab8e1f5375f85cc2d128b045ceff0460862ec9e8af5cf67820a6651e",0]
+["cp_async_mbarrier_arrive_shared",0,0,"sm_80","+ptx70","sm_80",[696,697,698,699,2],"4ec7d8c084e6678208bdbbddcc0d201f19576effff236a2d94258bcf059c8657",0]
+["cp_async_wait_all",0,0,"sm_80","+ptx70","sm_80",[700,701,702,703,2],"ee0b77a835ae39de82bae2e7832c64ed8e50b3dc4d7a89a6b45ddede4939554a",0]
+["cp_async_wait_group",0,0,"sm_80","+ptx70","sm_80",[704,705,706,707,2],"d37e5c7ddc901943f98e7d93e39cc2205eb29a1b6216208755e35f36b6cf26bf",0]
+["cvt_f16x2_f32",1,0,"sm_80","+ptx70","sm_80",[708,null,null,709,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_f32x2_bf16x2",1,0,"sm_80","+ptx70","sm_80",[710,null,null,711,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_f16x2_e4m3x2",1,0,"sm_89","+ptx81","sm_89",[712,null,null,713,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_f16x2_e5m2x2",1,0,"sm_89","+ptx81","sm_89",[715,null,null,716,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_relu_bf16x2_f32",1,0,"sm_80","+ptx70","sm_80",[717,null,null,718,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_relu_f16x2_e4m3x2",1,0,"sm_89","+ptx81","sm_89",[719,null,null,720,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_relu_f16x2_e5m2x2",1,0,"sm_89","+ptx81","sm_89",[721,null,null,722,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_relu_f16x2_f32",1,0,"sm_80","+ptx70","sm_80",[723,null,null,724,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_relu_satfinite_tf32_f32",0,0,"sm_100","+ptx86","sm_100",[725,726,727,728,21],"ab0a423e78569626f5ab97c09b78c28d4adcfed23f08109b65dfcee81c17c039",0]
+["cvt_rn_relu_tf32_f32",0,0,"sm_90","+ptx78","sm_90",[729,730,731,732,14],"d23b65afcd9de3224d763f046d5b783861e0a570815e9043e540c32b66ca3345",0]
+["cvt_rn_satfinite_e4m3x2_f16x2",1,0,"sm_89","+ptx81","sm_89",[733,null,null,734,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_e4m3x2_f32",1,0,"sm_89","+ptx81","sm_89",[735,738,736,737,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_e5m2x2_f16x2",1,0,"sm_89","+ptx81","sm_89",[739,null,null,740,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_e5m2x2_f32",1,0,"sm_89","+ptx81","sm_89",[741,744,742,743,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_relu_e4m3x2_f16x2",1,0,"sm_89","+ptx81","sm_89",[745,null,null,746,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_relu_e4m3x2_f32",1,0,"sm_89","+ptx81","sm_89",[747,750,748,749,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_relu_e5m2x2_f16x2",1,0,"sm_89","+ptx81","sm_89",[751,null,null,752,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_relu_e5m2x2_f32",1,0,"sm_89","+ptx81","sm_89",[753,756,754,755,714],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rn_satfinite_tf32_f32",0,0,"sm_100","+ptx86","sm_100",[757,758,759,760,21],"f43f8a9241cb7a013871f67cee705ba78d98772a3a7e79a1c607156c28ca7ce0",0]
+["cvt_rn_tf32_f32",0,0,"sm_90","+ptx78","sm_90",[761,762,763,764,14],"67c4b7c9bd2c661b4e068f8a54f9ae0df884bf27cfa5e8fd043a32109266fb3d",0]
+["cvt_rna_satfinite_tf32_f32",0,0,"sm_80","+ptx81","sm_80",[765,766,767,768,2],"5438d33b375195c2a2a11fd92e00f0b834b00a3c26bcc4148a65c388e078865b",0]
+["cvt_rna_tf32_f32",0,0,"sm_80","+ptx70","sm_80",[769,770,771,772,2],"bca414069c3c4faf3d1d4a6efce488a30fc66463bb0dbe8ec598a47805f0f0c0",0]
+["cvt_rz_bf16x2_f32",1,0,"sm_80","+ptx70","sm_80",[773,null,null,774,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rz_f16x2_f32",1,0,"sm_80","+ptx70","sm_80",[775,null,null,776,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["cvt_rz_relu_satfinite_tf32_f32",0,0,"sm_100","+ptx86","sm_100",[777,778,779,780,21],"0a97b884361963bf8655f6fdb5d8aea9eeaeb78d8ca1cd1e8d2c2a6ca71d87e5",0]
+["cvt_rz_relu_tf32_f32",0,0,"sm_90","+ptx78","sm_90",[781,782,783,784,14],"615cdfd60b0f66e5c0535f8ec354cf1ead392891f3b4214d527fbe627fff1bc2",0]
+["cvt_rz_satfinite_tf32_f32",0,0,"sm_100","+ptx86","sm_100",[785,786,787,788,21],"dc750fb19dc5c8674e4e4f3d4ca7e5a0574458047cc04499ea51c0361669ef75",0]
+["cvt_rz_tf32_f32",0,0,"sm_90","+ptx78","sm_90",[789,790,791,792,14],"2b9732d212b89a3c82c461b893838b42b1c6c045033f8720fcedf2b2c4681524",0]
+["div_rm_f32",1,0,"sm_80","+ptx70","sm_80",[793,796,794,795,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rm_f64",1,0,"sm_80","+ptx70","sm_80",[797,800,798,799,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rm_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[801,804,802,803,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rn_f32",1,0,"sm_80","+ptx70","sm_80",[805,808,806,807,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rn_f64",1,0,"sm_80","+ptx70","sm_80",[809,812,810,811,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rn_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[813,816,814,815,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rp_f32",1,0,"sm_80","+ptx70","sm_80",[817,820,818,819,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rp_f64",1,0,"sm_80","+ptx70","sm_80",[821,824,822,823,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rp_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[825,828,826,827,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rz_f32",1,0,"sm_80","+ptx70","sm_80",[829,832,830,831,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rz_f64",1,0,"sm_80","+ptx70","sm_80",[833,836,834,835,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["div_rz_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[837,840,838,839,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["dp2a_s32",0,0,"sm_61","+ptx50",null,[841,842,843,844,null],"b4fc2cb15e65bdd276f746d4c519140dff38491560c4d9f04c9194f2a4d290fb",0]
+["dp2a_s32",1,0,"sm_61","+ptx50","sm_75",[845,848,846,844,847],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["dp2a_u32",0,0,"sm_61","+ptx50",null,[849,850,851,852,null],"1e6772b9e969d70034340b03a7871ce9c3c0817c24439877a15b6bca2fe0e3d6",0]
+["dp2a_u32",1,0,"sm_61","+ptx50","sm_75",[853,855,854,852,847],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["dp4a_s32",0,0,"sm_61","+ptx50",null,[856,857,858,859,null],"923398e933aa61162ee09eb81c102559219d8f48b2cde55b4adb7fd94c293ef2",0]
+["dp4a_s32",1,0,"sm_61","+ptx50","sm_75",[860,862,861,859,847],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["dp4a_u32",0,0,"sm_61","+ptx50",null,[863,864,865,866,null],"d4e556a4320564c7c68f8995519aa4cfafdfde700cc9c796a8ec6495e2bdbdd6",0]
+["dp4a_u32",1,0,"sm_61","+ptx50","sm_75",[867,869,868,866,847],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["dsmem_read_u32",0,0,"sm_90","+ptx78","sm_90",[870,null,null,871,14],"4fa97e11de016a59ad2d5ba5bf3e90ee38d373b0d3635ba40984680b7abf07c4",0]
+["dynamic_smem_size",1,0,"sm_20","+ptx41",null,[872,null,null,873,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["dynamic_smem_size",1,0,"sm_20","+ptx41","sm_75",[872,null,null,873,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["elect_sync",1,0,"sm_90","+ptx80","sm_90",[874,877,875,876,14],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["envreg1",1,0,"sm_20","+ptx32",null,[878,881,879,880,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["envreg1",1,0,"sm_20","+ptx32","sm_75",[878,881,879,880,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["envreg2",1,0,"sm_20","+ptx32",null,[882,885,883,884,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["envreg2",1,0,"sm_20","+ptx32","sm_75",[882,885,883,884,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["ex2_approx_f16",1,0,"sm_75","+ptx70","sm_75",[886,null,null,887,7],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["ex2_approx_f32",1,0,"sm_80","+ptx70","sm_80",[888,null,null,889,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["ex2_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[890,null,null,891,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fence_mbarrier_init_release_cluster",0,0,"sm_90","+ptx80","sm_90",[892,null,null,893,14],"7b3f4224af427e221199a436d6dd5cb1934fcf15806d6bd5b35e8b1f62a71465",0]
+["fence_proxy_async_generic_acquire_shared_cluster_cluster",0,0,"sm_90","+ptx86","sm_90",[894,null,null,895,14],"ce13f0bb6abad1f28ff23990f6932204f01844557b6bcdfd72e28e2eda800abc",0]
+["fence_proxy_async_generic_release_shared_cta_cluster",0,0,"sm_90","+ptx86","sm_90",[896,null,null,897,14],"fd528b8fba2016301322b4c63c14c1f057b3d18547468d8bf69455cf547e408d",0]
+["fence_proxy_async_shared_cta",0,0,"sm_90","+ptx80","sm_90",[898,null,null,899,14],"b7ffef1852f0170e5f317ca6c785ce171792fb3376e6c6f354be6c88453a12a8",0]
+["fence_proxy_tensormap_generic_acquire_cluster",0,0,"sm_90","+ptx83","sm_90",[900,901,902,903,14],"eaa1b71fb31880f0e3f462a919d5f47c56f30aad004e26b9d138d372adad4382",0]
+["fence_proxy_tensormap_generic_acquire_cta",0,0,"sm_90","+ptx83","sm_90",[904,905,906,907,14],"8d31b4dcae3d1646e2f541ce900d240b23397b7665ac357c85f5fad06cae8e51",0]
+["fence_proxy_tensormap_generic_acquire_gpu",0,0,"sm_90","+ptx83","sm_90",[908,909,910,911,14],"58d7168791b44e23f1b07dc148630a178eb1a50400220f99c10767789ed049fb",0]
+["fence_proxy_tensormap_generic_acquire_system",0,0,"sm_90","+ptx83","sm_90",[912,913,914,915,14],"c4975f751c8a642ef5517866a4eab181000b3a0164d1f5000347d1445df3b362",0]
+["fence_proxy_tensormap_generic_release_cluster",0,0,"sm_90","+ptx83","sm_90",[916,917,918,919,14],"51c792a551a2339b34893e589cfb03992ee4b8ae7881c7af288495f1160268f1",0]
+["fence_proxy_tensormap_generic_release_cta",0,0,"sm_90","+ptx83","sm_90",[920,921,922,923,14],"473ae923f46108f1084ad2988f30a4089341926a3de8abaf8e1af51790215df2",0]
+["fence_proxy_tensormap_generic_release_gpu",0,0,"sm_90","+ptx83","sm_90",[924,925,926,927,14],"a56a11477f3b23ee1774e676219586d23b93911880847becc9f1e52cc0fb465f",0]
+["fence_proxy_tensormap_generic_release_system",0,0,"sm_90","+ptx83","sm_90",[928,929,930,931,14],"3b030a42f4994773aecef73875722047e6e9677370a2355f47c07c6055520457",0]
+["fma_bf16x2",0,0,"sm_80","+ptx70","sm_80",[932,null,null,933,2],"58024626830ac6fd88d43c20ad8d6b3bdc29ee0b0e7c3b9f234fbb71dc76adf0",0]
+["fma_f16x2",0,0,"sm_70","+ptx60",null,[934,null,null,935,null],"997c420a90563e7d9e98f251663e5e4ecaa1903c8da50ee184e807769c63a000",0]
+["fma_f16x2",1,0,"sm_70","+ptx60","sm_75",[936,null,null,935,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["fma_f32x2",0,0,"sm_100","+ptx86","sm_100",[937,null,null,938,21],"6a4ea4ba97cde9c543c78ea379a875dacab4ec993c6a412e5ab4251d0b5403d3",0]
+["fma_ftz_f16x2",0,0,"sm_70","+ptx60",null,[939,null,null,940,null],"e71e09811683e1bc4ce7308d199468cf2366fe8b16d2fd6f50cb9a8b60d21677",0]
+["fma_ftz_f16x2",1,0,"sm_70","+ptx60","sm_75",[941,null,null,940,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["fma_ftz_f32x2",0,0,"sm_100","+ptx86","sm_100",[942,null,null,943,21],"ac313e89a8ca6cc358d1e0ea314e7d423bf6b4ad2b0fa38ec6d23c6952fc2fad",0]
+["fma_ftz_relu_f16x2",0,0,"sm_80","+ptx70","sm_80",[944,null,null,945,2],"49ae7152e43e112b2dd212e5fcd23518dcd885191b736fcda14bc5398c7bb3ca",0]
+["fma_ftz_sat_f16x2",0,0,"sm_70","+ptx60",null,[946,null,null,947,null],"d12899702473ab67053b925dbd5555256f0cefb5b626d1c45ed41b3037d1ebc3",0]
+["fma_ftz_sat_f16x2",1,0,"sm_70","+ptx60","sm_75",[948,null,null,947,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["fma_relu_bf16x2",0,0,"sm_80","+ptx70","sm_80",[949,null,null,950,2],"b5d963e414b558564eecf620b1872a20b5a991e901121fbc8cf7417a4c4e39ac",0]
+["fma_relu_f16x2",0,0,"sm_80","+ptx70","sm_80",[951,null,null,952,2],"1efd8e1fc9d0e4945fa9db901747a6f0554fb3fce3c754cafb4fb489323414f7",0]
+["fma_rm_f32",1,0,"sm_80","+ptx70","sm_80",[953,956,954,955,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rm_f64",1,0,"sm_80","+ptx70","sm_80",[957,960,958,959,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rm_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[961,964,962,963,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rm_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[965,null,null,966,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rm_sat_f32",1,0,"sm_80","+ptx70","sm_80",[967,null,null,968,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rn_f32",1,0,"sm_80","+ptx70","sm_80",[969,972,970,971,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rn_f64",1,0,"sm_80","+ptx70","sm_80",[973,976,974,975,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rn_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[977,980,978,979,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rn_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[981,null,null,982,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rn_sat_f32",1,0,"sm_80","+ptx70","sm_80",[983,null,null,984,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rp_f32",1,0,"sm_80","+ptx70","sm_80",[985,988,986,987,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rp_f64",1,0,"sm_80","+ptx70","sm_80",[989,992,990,991,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rp_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[993,996,994,995,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rp_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[997,null,null,998,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rp_sat_f32",1,0,"sm_80","+ptx70","sm_80",[999,null,null,1000,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rz_f32",1,0,"sm_80","+ptx70","sm_80",[1001,1004,1002,1003,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rz_f64",1,0,"sm_80","+ptx70","sm_80",[1005,1008,1006,1007,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rz_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[1009,1012,1010,1011,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rz_ftz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[1013,null,null,1014,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_rz_sat_f32",1,0,"sm_80","+ptx70","sm_80",[1015,null,null,1016,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["fma_sat_f16x2",0,0,"sm_70","+ptx60",null,[1017,null,null,1018,null],"25beba7af20d1853177742e68523cb9a98f21e32dfecf1c696ebc170a6c2d8fd",0]
+["fma_sat_f16x2",1,0,"sm_70","+ptx60","sm_75",[1019,null,null,1018,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["globaltimer",1,0,"sm_30","+ptx32",null,[1020,1023,1021,1022,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["globaltimer",1,0,"sm_30","+ptx32","sm_75",[1020,1023,1021,1022,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["grid_dependency_launch_dependents",0,0,"sm_90","+ptx78","sm_90",[1024,1025,1026,1027,14],"85a5bedf6122f942514b646410360cddfbeb69cb8a167190373971acb85f06e8",0]
+["grid_dependency_wait",0,0,"sm_90","+ptx78","sm_90",[1028,1029,1030,1031,14],"14479222c064740878f292bbda8944aa0e6b362332e80a8926086cee5239d166",0]
+["grid_dim_x",0,0,"sm_70","+ptx60",null,[1032,1033,1034,1035,null],"7874bce13275cea505847cacb4ef98acd0f0fb7a22cce8e731c39708b45eed77",0]
+["grid_dim_y",0,0,"sm_70","+ptx60",null,[1036,1037,1038,1039,null],"9cdbba347422496f955ac8707823ef564020fcb0d429f34ae237e28f1cd9831f",0]
+["grid_dim_z",0,0,"sm_70","+ptx60",null,[1040,1041,1042,1043,null],"f613973309c2a67694be8c083405039530ec444cddebab5ce60a8b6e436ad21c",0]
+["gridid",1,0,"sm_30","+ptx32",null,[1044,null,null,1045,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["gridid",1,0,"sm_30","+ptx32","sm_75",[1044,null,null,1045,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["lane_id",0,0,"sm_70","+ptx60",null,[1046,1047,1048,1049,null],"2258f9ef982bb36dbf3033ad70970a30b41485467a4e562d674955ecbefa9700",0]
+["lanemask_eq",0,0,"sm_20","+ptx32",null,[1050,1051,1052,1053,null],"c73bf54220d4d5159daadff11e1b976a83bf6a2b9c3098859f5af01b926fa910",0]
+["lanemask_ge",0,0,"sm_20","+ptx32",null,[1054,1055,1056,1057,null],"b428439b8f988ed139d3366160c49910273c49aa4744b1bce487ca3f815688ac",0]
+["lanemask_gt",0,0,"sm_20","+ptx32",null,[1058,1059,1060,1061,null],"1348f24d5cdb9669a69553c784a198a0fbfc89911eeaa201b614121493a97448",0]
+["lanemask_le",0,0,"sm_20","+ptx32",null,[1062,1063,1064,1065,null],"d3f8174daf79e717fb7e5dd411fcd6fc26736ae9f6ef79cdac3799be4ed4af97",0]
+["lanemask_lt",0,0,"sm_20","+ptx32",null,[1066,1067,1068,1069,null],"c66b64e0355f43cd58252900c12e8c1219610915242a7b031c675b0890ba6e82",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_100a","+ptx86","sm_100a",[1070,1071,1072,1073,213],"f26febf0f4117a4599a7bf842e8e0272a2ca40526002ab69007216714faf2feb",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_100f","+ptx88","sm_100f",[1074,1071,1072,1075,1076],"7f6f5db4712e53867d0dd7928b88540dd961b467ea47618e6ce05e0278e134e7",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_103a","+ptx88","sm_103a",[1077,1071,1072,1078,1079],"363f8845c7de85b105732abffe878dbb6361b216a49befc1f6f9654e99515c74",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_103f","+ptx88","sm_103f",[1080,1071,1072,1081,1082],"20a1ff6851b4d2e847171d1150382c848b2867c1e8b6abbb35cff972ea4f88e0",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_110a","+ptx90","sm_110a",[1083,1071,1072,1084,1085],"7cb33cc19b40e0bc9061289133081b9f43c7d379a597648f1b7dbd3c86b56eec",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_110f","+ptx90","sm_110f",[1086,1071,1072,1087,1088],"919eed202f9de49e65c8a3f98f11db92579c33843166b8156d1007f6bb049882",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_120a","+ptx87","sm_120a",[1089,1071,1072,1090,1091],"afcc0c0e2abe286c22806d3b072631bfa100c0a4c4432078d2de933d1b9d7c55",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_120f","+ptx88","sm_120f",[1092,1071,1072,1093,1094],"e901b48dbfeb6aae7f2f48ac33271932944587f96b49828f28a33b66f0e48ed5",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_121a","+ptx88","sm_121a",[1095,1071,1072,1096,1097],"3f2cd6b948014ca91c9a3191eafc2dba8c4e36393941386f3ab4a1da5c12d824",0]
+["ldmatrix_m16n16_x1_trans_b8",0,0,"sm_121f","+ptx88","sm_121f",[1098,1071,1072,1099,1100],"e4f615c05d2a78f45889afcdb0679a1383d5465f35021aaa2cf874cfa79064e9",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[1101,1102,1103,1104,213],"faee871d1db4ee079e1f098270e3ecea8434ebe506f68b130cd4e30669786fbf",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_100f","+ptx88","sm_100f",[1105,1102,1103,1106,1076],"82deaa0f4f828a46fefaaf83b6c4847fa8d3ed266298278cfcb703b85b8ab046",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_103a","+ptx88","sm_103a",[1107,1102,1103,1108,1079],"de748e12998a81c5e4bf9d0f07c309f8e3b5287aecd6bd88213c06e6ae17d4e1",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_103f","+ptx88","sm_103f",[1109,1102,1103,1110,1082],"1bfd30ceb584ce9bde6d0a75befc2e94630e75553a251462225990d431ed8a50",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_110a","+ptx90","sm_110a",[1111,1102,1103,1112,1085],"5549416aa9378d14e6314a4ac6893e0a55e433c1bd68c1112fc0380cea8ad97d",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_110f","+ptx90","sm_110f",[1113,1102,1103,1114,1088],"aacbf74f07475bbe82160f7747882b4132c9e899f59617d86c6a91272113f4e6",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_120a","+ptx87","sm_120a",[1115,1102,1103,1116,1091],"5eab509810349e863444f46736fae5ac12c89126ee6be456b824e90eb60b3f3e",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_120f","+ptx88","sm_120f",[1117,1102,1103,1118,1094],"2811c38d4bbb1f7ced069b2b0032aa2a2a8ada12abfc504d0edbd88935bcd5fd",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_121a","+ptx88","sm_121a",[1119,1102,1103,1120,1097],"61d991ce9002569c9f6a97780cd70a6e63c282b21700165564f8407af2338e9e",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64",0,0,"sm_121f","+ptx88","sm_121f",[1121,1102,1103,1122,1100],"3b8e1a60a80f78fc70d543df675a60cd80856e584ff786ebc69b27b47e37d518",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[1123,1124,1125,1126,213],"1d41c2c5501142dfb1707dc3388d25658e859b30f44c7bee688204d8b28c74e8",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_100f","+ptx88","sm_100f",[1127,1124,1125,1128,1076],"9243a4a50bd5dd5a4031f965834823dff862106b33454ea6405650d961933443",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_103a","+ptx88","sm_103a",[1129,1124,1125,1130,1079],"b7f15632bdd2ad3b23f6a1c18cd71bc0fdbd08e0b537fdfdfed6c652f00ab2b6",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_103f","+ptx88","sm_103f",[1131,1124,1125,1132,1082],"686d359944f510619e068167244fc1d00f34db7981a515446ee1cdecd131ce55",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_110a","+ptx90","sm_110a",[1133,1124,1125,1134,1085],"ba6bfc5f4afd29f878c4173b4b6ef819fc11430c25093216f509c7178bc67122",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_110f","+ptx90","sm_110f",[1135,1124,1125,1136,1088],"a953bc69d6cfc72ef4934758fc6dc5434c92d836de9ce463f0647621bc0a0edc",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_120a","+ptx87","sm_120a",[1137,1124,1125,1138,1091],"31718aef9418d79567a5baaebe8b477e16c43fa95bfdeac92d5ae7b88ac7e1cd",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_120f","+ptx88","sm_120f",[1139,1124,1125,1140,1094],"df9e290592b2217275cbcccdbc487eefe124d691d0381712902ee61719a19471",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_121a","+ptx88","sm_121a",[1141,1124,1125,1142,1097],"717ea7994fd24f024044eeebacdb0faa2623ee4fbcc083092ca5353da7cfa034",0]
+["ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32",0,0,"sm_121f","+ptx88","sm_121f",[1143,1124,1125,1144,1100],"278de1f06d12a36a78035244b54b9dc265d466b25221d282d004b79256fca1bf",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_100a","+ptx86","sm_100a",[1145,1146,1147,1148,213],"264b672a7ce8208261b7b9b7bea12a16bec02ce21ccd66586989747b30119cfe",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_100f","+ptx88","sm_100f",[1149,1146,1147,1150,1076],"663ae9c7ed454fab911cab0e1e035e0ffc5c4748aa3d7142fb5aaac9ebca061b",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_103a","+ptx88","sm_103a",[1151,1146,1147,1152,1079],"740b569692a82137bc2d5d3ef2714ba3ae37b75ff3f55d164049b29ab8d11491",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_103f","+ptx88","sm_103f",[1153,1146,1147,1154,1082],"7cae2daa79d20fab898b758c93913b17cf904b5dcef58b17d9ac496e5f67ee1a",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_110a","+ptx90","sm_110a",[1155,1146,1147,1156,1085],"6c9c40013a682de4ca4838dbd0b439fc348ee61d54818286773927c72bce5850",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_110f","+ptx90","sm_110f",[1157,1146,1147,1158,1088],"415cb35b9f89140ebfe59c42679e3ec857aea978bc22dfce305d7a9e9775aa7d",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_120a","+ptx87","sm_120a",[1159,1146,1147,1160,1091],"036fbb4edfa5cdccf4c29bb827233c0ffaba5fe1128c22307bcb48359475c65b",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_120f","+ptx88","sm_120f",[1161,1146,1147,1162,1094],"dd179c355ae8a58a7a46b2adad7ba1a015f58d65cb1585789658b83e76c08844",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_121a","+ptx88","sm_121a",[1163,1146,1147,1164,1097],"a9a2942f14027b44bbfae509de6c4cc2c48488893c955d5f06f0b5076d4ef354",0]
+["ldmatrix_m16n16_x2_trans_b8",0,0,"sm_121f","+ptx88","sm_121f",[1165,1146,1147,1166,1100],"a92c6544cdd293580635e09655b60efc6f8c0727108577de47396326806ef831",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[1167,1168,1169,1170,213],"3b7a7edcff83e5be8e3bbb8a4431f6ead49bbc63e08c22c6d0b76bf22daeafd8",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_100f","+ptx88","sm_100f",[1171,1168,1169,1172,1076],"94235cea0845659cb0088b1d9da5fee5ac54550fefc655fe745099ad5ddafb17",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_103a","+ptx88","sm_103a",[1173,1168,1169,1174,1079],"019039f4497b771ffda80770611b6753e126d44f6bdaeb0904865808a9143625",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_103f","+ptx88","sm_103f",[1175,1168,1169,1176,1082],"9b946a160096f4ce33f56619cb344b7e9f19709c09f145c02a20906849e08640",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_110a","+ptx90","sm_110a",[1177,1168,1169,1178,1085],"e7fdad9bb49533f02733172dc5e3027f68f88258f58e7eeaebdb8fb8a25b96e3",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_110f","+ptx90","sm_110f",[1179,1168,1169,1180,1088],"c989ebf16013a0c1c6a8c9e0d679dd609acf0eed0bc00865e9d6226f7e88086a",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_120a","+ptx87","sm_120a",[1181,1168,1169,1182,1091],"79ada238721a2bc104366b31462ef7b2560102cd7a28f770f337395e65e5d43f",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_120f","+ptx88","sm_120f",[1183,1168,1169,1184,1094],"6888384a55c19ce0accf40eafb380ff6592deb8d6d4324946a126068a20998f4",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_121a","+ptx88","sm_121a",[1185,1168,1169,1186,1097],"54b43dd0db367cbbb831fc355ef11742376353df6709e83e7e4edb76d6e83a42",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64",0,0,"sm_121f","+ptx88","sm_121f",[1187,1168,1169,1188,1100],"e02a24540c0034e5d16cdd878afec505f5e621e49124b6dfc725c337ad483f0a",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[1189,1190,1191,1192,213],"0974b0e0e3ac257b79a6f397bd3a8d6f2edab4d8add538d9f3e66e1e56589c5f",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_100f","+ptx88","sm_100f",[1193,1190,1191,1194,1076],"f762e4cf9d74fb9ab524a5c18971c0a01d0fbd1e1e48b53900503b6a6c93bbc0",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_103a","+ptx88","sm_103a",[1195,1190,1191,1196,1079],"fadd73697519546340dee9456972a83e22ea9cd780499d584bc7196cd40360c4",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_103f","+ptx88","sm_103f",[1197,1190,1191,1198,1082],"9216154013b0c58369a8ede45d92c70b199c30b2d13083d74eedcb0525978597",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_110a","+ptx90","sm_110a",[1199,1190,1191,1200,1085],"d414bcb4de05276014c1a01e955608b7e1fe86003233d54cb486f19e034eafbd",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_110f","+ptx90","sm_110f",[1201,1190,1191,1202,1088],"be404179d46d5368bfdf0890dbcd702e9dee8e7cd6827eac9bfd12fd6560dab0",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_120a","+ptx87","sm_120a",[1203,1190,1191,1204,1091],"dcc24a92cd5f11243a72d51aa3a31ce0b1eb31733bed466414c0389b7b966e02",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_120f","+ptx88","sm_120f",[1205,1190,1191,1206,1094],"6bc098c27c025a391f9a5da7f9ed7d82c3267a9c90f4ed7c83befb0463071c8c",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_121a","+ptx88","sm_121a",[1207,1190,1191,1208,1097],"a6223fe7cb5c21076e35d9f7e0c37013472015e77616589fc6892e00a213cb90",0]
+["ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32",0,0,"sm_121f","+ptx88","sm_121f",[1209,1190,1191,1210,1100],"4c672eff003a36bb454eda4c3ba9aa09fcda42ae693a117cbaf049138b92c630",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[1211,1212,1213,1214,213],"eb952ee0a8d5c7e7268c6a5c826705143487171fb376a6a894262751d77e2ce7",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_100f","+ptx88","sm_100f",[1215,1212,1213,1216,1076],"bbafbd3a365d4a128aef318edc8ac8bb624a143dccfe2c606b5e255b496e8851",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_103a","+ptx88","sm_103a",[1217,1212,1213,1218,1079],"f48e8bbf258fb8a99150ae761823b1d201148bb774e11e97f8f09601ee968efc",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_103f","+ptx88","sm_103f",[1219,1212,1213,1220,1082],"58e664acbbb96c0e72396b80e26fd78e8f907a87e745010e0fcb9e95c4e7a787",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_110a","+ptx90","sm_110a",[1221,1212,1213,1222,1085],"9162516d43de7b9d912704eac3f3e7973e4d12f3d6e5b603075334d47c404a3e",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_110f","+ptx90","sm_110f",[1223,1212,1213,1224,1088],"94de4f20130a187a18d1117ff4828374b2fdc15370023508dd700e177a4673ea",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_120a","+ptx87","sm_120a",[1225,1212,1213,1226,1091],"2abf23b2d255ae80c7ac5bd1b73aa688458143a5dad8bb905194e2c56a0e8219",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_120f","+ptx88","sm_120f",[1227,1212,1213,1228,1094],"0f920eca898367bf4c41a3fc26aa39ee8c60f7a929e215461e5705f0df14efd1",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_121a","+ptx88","sm_121a",[1229,1212,1213,1230,1097],"d42cbb14755224136355482af90a12e6fd2a477b03660863e1781c634e995a3e",0]
+["ldmatrix_m8n16_x1_b8x16_b4x16_p64",0,0,"sm_121f","+ptx88","sm_121f",[1231,1212,1213,1232,1100],"a5f3ce0c5ca6ebf4ded3e5d4b6f398da216f153242c2f729fd5fb397e7e778dd",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[1233,1234,1235,1236,213],"53f0b637fdbcced7d97ec74a080f7dbd40baca6b59639cc2cdbf5463c411d9cd",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_100f","+ptx88","sm_100f",[1237,1234,1235,1238,1076],"37a11791c9c55e9f90569ea025d7ddb39dd870af1b8543546490d98d55aec338",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_103a","+ptx88","sm_103a",[1239,1234,1235,1240,1079],"11907dd06bcfb71370801b0a2778f58a763eea9e5aa0c7574bd336967d5a7dd1",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_103f","+ptx88","sm_103f",[1241,1234,1235,1242,1082],"d94b3a6a1af9db3c294fd894263a7c7739e035f01a5e2b0a3be335d0e0c2c72a",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_110a","+ptx90","sm_110a",[1243,1234,1235,1244,1085],"e714cd460cbe0d79246454f81e9d339b888e87c6b5a208269cea10b7a40e07d6",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_110f","+ptx90","sm_110f",[1245,1234,1235,1246,1088],"f2c8705eae8e7a85b144015742756cebdb9522623180e48824acaefb99792023",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_120a","+ptx87","sm_120a",[1247,1234,1235,1248,1091],"577c60d434690ae2e40f611f40d8de0b6ddfb5a6d7854f769c1432c591eb8a88",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_120f","+ptx88","sm_120f",[1249,1234,1235,1250,1094],"2c1adeec00f9f818743a89b5eb9cd38478f2c3e9057ba5cddc5bd362d77ebd3f",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_121a","+ptx88","sm_121a",[1251,1234,1235,1252,1097],"80f954c939e4ec2a193f0761dd2268eb69e928b0ff67f729c2eddd5ae9f67004",0]
+["ldmatrix_m8n16_x1_b8x16_b6x16_p32",0,0,"sm_121f","+ptx88","sm_121f",[1253,1234,1235,1254,1100],"5a903cf0e7d7d32eb52f6c5d08e5df25d18d348d34525a3aaae003476d0aa169",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[1255,1256,1257,1258,213],"c362da04894b3035f6247d99a1175de96840a40ad5cbc44e05eb62bec1ef5c49",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_100f","+ptx88","sm_100f",[1259,1256,1257,1260,1076],"b6580291f372e856958135af91fc92d806ba86b8c7697fb6d7f838c5ec407618",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_103a","+ptx88","sm_103a",[1261,1256,1257,1262,1079],"b44090123b541274fe03e0d08347ebb2e7000cf60363c6084d2a1f6eb85cc41a",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_103f","+ptx88","sm_103f",[1263,1256,1257,1264,1082],"47762ea2d3845be8c227e9fc280948e4f9a5b121eafc66f53b1b5b50c1f91129",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_110a","+ptx90","sm_110a",[1265,1256,1257,1266,1085],"6a8ad60c0bba7d4b59dfdbed23a214b1ae082157ec0a161c2d95e80387bad32f",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_110f","+ptx90","sm_110f",[1267,1256,1257,1268,1088],"65b8df922000fb174e64975101cf3fb132cc597be22632abb2e5a69a1b86849e",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_120a","+ptx87","sm_120a",[1269,1256,1257,1270,1091],"ffd9cf5a995f96c52a27eccf5124260205e3d0f5cd1f15ac149879a35ed7ca33",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_120f","+ptx88","sm_120f",[1271,1256,1257,1272,1094],"eafa45b88dba51aa48a859decfd1c7ad44dd7be1d71500c20e8d9d564f890c62",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_121a","+ptx88","sm_121a",[1273,1256,1257,1274,1097],"11648e25cad5966081746565e0a54ef136871bfa2441be36e1597b5b9faa2a67",0]
+["ldmatrix_m8n16_x2_b8x16_b4x16_p64",0,0,"sm_121f","+ptx88","sm_121f",[1275,1256,1257,1276,1100],"692edb8a8b1fd73671079388ff0b7821218a4ecae8bb4496717632695abfe4e1",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[1277,1278,1279,1280,213],"dec1f09bac8ca41ceab9e5e31389f755a24bf606551c88e6d89f14a7b8805d0b",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_100f","+ptx88","sm_100f",[1281,1278,1279,1282,1076],"1ce1f9ee602976e38a4f2935ace39a8bcc12acb708fdbd774c0cfe2edcae1cbc",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_103a","+ptx88","sm_103a",[1283,1278,1279,1284,1079],"aece688e2c82b67a10dc378695b166faa47a629d8485733b75f9515120f4ed6c",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_103f","+ptx88","sm_103f",[1285,1278,1279,1286,1082],"5fec1119799a0fa21e93edf735b607ada57e89b616be1bf02dec94341c9a9520",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_110a","+ptx90","sm_110a",[1287,1278,1279,1288,1085],"bcad0bd1f59284f24a822a209d1999c553ab89b94ef0000909dade5ca443b97c",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_110f","+ptx90","sm_110f",[1289,1278,1279,1290,1088],"837b4bb50197ee05d836396aaf007a835abed8c2552dc1f69010a523402242c4",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_120a","+ptx87","sm_120a",[1291,1278,1279,1292,1091],"b733b78e72b240466a5a44c9c3de66b850e74c0ade357e50bb7b2fd40e224db8",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_120f","+ptx88","sm_120f",[1293,1278,1279,1294,1094],"31ce0c3bc0dd2e607210550edacbac0c9d90f2b85bb9c918e76d12d08e0fcf6e",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_121a","+ptx88","sm_121a",[1295,1278,1279,1296,1097],"81a4ab975ee2eab2b114443e82abd2de8975938e76d87365f48a6d2fd8e89fa1",0]
+["ldmatrix_m8n16_x2_b8x16_b6x16_p32",0,0,"sm_121f","+ptx88","sm_121f",[1297,1278,1279,1298,1100],"c503263d34a923a290860e16adfce2204bd81ca4b2e1b97cd5f5f4fc510958e8",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[1299,1300,1301,1302,213],"49b0d4b6a23bd84039304cbb46bfbe1ba1dc5e523fde775be238c4f52f8c17e6",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_100f","+ptx88","sm_100f",[1303,1300,1301,1304,1076],"ebcbcea4310852cbf4e8116d761e83a740bbe79fb1da4236633e0ec0f7cc1f6f",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_103a","+ptx88","sm_103a",[1305,1300,1301,1306,1079],"33cceb0b350364d64b37e4a958f8e7cb016b8b4580773f054da3e1a03bdd5fe8",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_103f","+ptx88","sm_103f",[1307,1300,1301,1308,1082],"a12c28e3691e5d47c66d560a73026a5f30afe8e11c7e9b0b187d973354f0ce8b",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_110a","+ptx90","sm_110a",[1309,1300,1301,1310,1085],"29ea0b7597ea4a07b10b1e9d76655f2417f44c1da3b8b390997d7b12b6226362",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_110f","+ptx90","sm_110f",[1311,1300,1301,1312,1088],"47fb0670fd246ebcd224200d934141babde70a9e74216c60c8d34762bfbfe8e6",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_120a","+ptx87","sm_120a",[1313,1300,1301,1314,1091],"8889d901f7ac2b2acff0b260fdefb0b98214dec13a672cf7aa1c05f41019cf1f",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_120f","+ptx88","sm_120f",[1315,1300,1301,1316,1094],"8246783b208a3d33e784442e3d528ca46fd3a49dc01efcb563019b20edd2c5f2",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_121a","+ptx88","sm_121a",[1317,1300,1301,1318,1097],"9e0beeadbb02cdf9d7fe786eafb769b1ca830054890d1f6d1201b91aa057255a",0]
+["ldmatrix_m8n16_x4_b8x16_b4x16_p64",0,0,"sm_121f","+ptx88","sm_121f",[1319,1300,1301,1320,1100],"038a077b7598663f194b7af188e4eab352a1e9b58e870e31f24ee34fb67fe46a",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[1321,1322,1323,1324,213],"35c5a51e58a8cb0f91863b4e1b216ce985dce79906494e2f46ccf2d7a06f5e32",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_100f","+ptx88","sm_100f",[1325,1322,1323,1326,1076],"c5b04bf11c8fb89d4feccec96a29ac228138cae0445aaca559f34d57b43a48df",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_103a","+ptx88","sm_103a",[1327,1322,1323,1328,1079],"841a2c54afd7453f19eb4370f9266046f58e3b87442c6058d89b31fad1e15185",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_103f","+ptx88","sm_103f",[1329,1322,1323,1330,1082],"4962a303dcafc3e43c374d73b3c3ac3e183e1face252c0f2cc6ffd39ff8c7631",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_110a","+ptx90","sm_110a",[1331,1322,1323,1332,1085],"70d617c7e0aa449d3fcf0dcfd6da4a1b163bc38a466f48a4cce6cedb68aed01d",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_110f","+ptx90","sm_110f",[1333,1322,1323,1334,1088],"6f1dcdbc5bd1a4de8b063b73c98740c501b91c3c4f4c2a12577d686d1369a01f",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_120a","+ptx87","sm_120a",[1335,1322,1323,1336,1091],"0a6a84439ee3bf6fc02152f8fbdabed6a7821a48a38d5fa0a14f3beb522aa5d1",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_120f","+ptx88","sm_120f",[1337,1322,1323,1338,1094],"9c6bf7a0a5728fc3eb38c9ee665d51a9987878387a4ea5e6294f6b5ecbf85162",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_121a","+ptx88","sm_121a",[1339,1322,1323,1340,1097],"5cb36c31c1d5477204b64b721e898f507eab7168efe3b129257b704d739fe6b2",0]
+["ldmatrix_m8n16_x4_b8x16_b6x16_p32",0,0,"sm_121f","+ptx88","sm_121f",[1341,1322,1323,1342,1100],"3e38c05c311333ec2f61dbc443c9d75c2b9dbc54ce2c797fe50de06d5285896a",0]
+["ldmatrix_m8n8_x1_b16",0,0,"sm_75","+ptx65","sm_75",[1343,1344,1345,1346,7],"7113304628044e69226791270391bf67fdb9accfe0badcdded05bea570b842f3",0]
+["ldmatrix_m8n8_x1_trans_b16",0,0,"sm_75","+ptx65","sm_75",[1347,1348,1349,1350,7],"c186c99a92a1325ea93ac0d211cbfb6623421136dba4bb69d26ff27c86afc35a",0]
+["ldmatrix_m8n8_x2_b16",0,0,"sm_75","+ptx65","sm_75",[1351,1352,1353,1354,7],"b497ac21fe5cfe7f8c9268433a04715cccd026ed9194539d74f18a57062bc8dc",0]
+["ldmatrix_m8n8_x2_trans_b16",0,0,"sm_75","+ptx65","sm_75",[1355,1356,1357,1358,7],"4c4836d7901b9e9c024a7c1d28c733bce1593273273c54b4b1ce69b30fb61828",0]
+["ldmatrix_m8n8_x4_b16",0,0,"sm_75","+ptx65","sm_75",[1359,1360,1361,1362,7],"de9ca481b0ddef485a397deccfb13b441e4cb5e467c3b110a33374ad3b9c8d57",0]
+["ldmatrix_m8n8_x4_trans_b16",0,0,"sm_75","+ptx65","sm_75",[1363,1364,1365,1366,7],"196946aeee27da1dc039411c1fbcc00d124617f10be57508305db18114ad8f3e",0]
+["lg2_approx_f32",1,0,"sm_80","+ptx70","sm_80",[1367,null,null,1368,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["lg2_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[1369,null,null,1370,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["map_shared_rank",0,0,"sm_90","+ptx78","sm_90",[1371,null,null,1372,14],"974d63200c95b6d18bdc05300645dc9b375ec64a7a3a582ff91da956f58a41bc",0]
+["match_all_i64_sync",0,0,"sm_70","+ptx60",null,[1373,1374,1375,1376,null],"65942a54f8507365e95f995e88252a7d0afc556d2fddfd6eb5909c60de5ef235",0]
+["match_all_sync",0,0,"sm_70","+ptx60",null,[1377,1378,1379,1380,null],"ca32b4e6df6bb245cb7fadac1f1146f55b13d8e66431240fc2e43032d72945f1",0]
+["match_any_i64_sync",0,0,"sm_70","+ptx60",null,[1381,1382,1383,1384,null],"bdacd2dc459dcfb5212656cfb3d2e452f099aa1220e9315ffabb8c71178750b5",0]
+["match_any_sync",0,0,"sm_70","+ptx60",null,[1385,1386,1387,1388,null],"5a2ab6bb9ab56e0a1cd67e9783ba8d623eb2f32181e96158717854d33c03a9d9",0]
+["max_bf16",0,0,"sm_80","+ptx70","sm_80",[1389,null,null,1390,2],"8ec9927d5aee6e5f39abfb8d3e6bd3462b2a066c0b4ee425a31c2342ffca753f",0]
+["max_bf16x2",0,0,"sm_80","+ptx70","sm_80",[1391,null,null,1392,2],"8bed15f5b1c93c6056da2fb6d5d30157e9a92a28b7d3bb2d104fe31da498cf90",0]
+["max_f16",0,0,"sm_80","+ptx70","sm_80",[1393,null,null,1394,2],"3686a7cd0e6cabf6c71e72d03454a7acec10ca34c66fa76f656d2f121b4d5e1b",0]
+["max_f16x2",0,0,"sm_80","+ptx70","sm_80",[1395,null,null,1396,2],"40e7da1ff0ae0c8a3996e6026d54b399de0eac5f19b91a2d0da6173a5b9e8fec",0]
+["max_ftz_f16",0,0,"sm_80","+ptx70","sm_80",[1397,null,null,1398,2],"15cc976552f4b7fe991196d6f8adc6c1ad231295b146eda5665fb1a0faa080d0",0]
+["max_ftz_f16x2",0,0,"sm_80","+ptx70","sm_80",[1399,null,null,1400,2],"a49163e42fad5fa10d4ab0cb329bec8c551d813e8d00e1d47656f4ebed55445a",0]
+["max_ftz_nan_f16",0,0,"sm_80","+ptx70","sm_80",[1401,null,null,1402,2],"b578d4ff00e1bb1d2a9d3fad26319b556b96ca41884d6d53b9c63f80d78ad8d6",0]
+["max_ftz_nan_f16x2",0,0,"sm_80","+ptx70","sm_80",[1403,null,null,1404,2],"76069824a3b33e1b9c518ea0fd837e437839a8819d2891b98b8b0325b74e01db",0]
+["max_ftz_nan_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1405,null,null,1406,1407],"e7ec8cdc7e44c9e6e98a7340aed725a8383a2df3621d3a77ddc8cb8880b5a052",0]
+["max_ftz_nan_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1408,null,null,1409,null],"568739697effdcc3238f6bf6e725e1cadd1b7479c861c7bbf9ae6748f77ea2e7",0]
+["max_ftz_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1410,null,null,1411,1407],"267bf981be9894d2c01d92da2f951df6ac9d77aeef36cdde10daa606f4d9977b",0]
+["max_ftz_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1412,null,null,1413,null],"c37d5a68dacc48f7644e8981defcbc7ff855dd4024afa3f1dd8bd8ea0ea1b664",0]
+["max_ftz_nan_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1414,null,null,1415,1407],"8fbe6b54dce8eb0b215e56a087b8b65e7ffa727097801e1c2e36793e544f61ce",0]
+["max_ftz_nan_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1416,null,null,1417,null],"df49861605025d07ec16d55893b73672278ad02c09c3ab8580c0fe261bfd72f9",0]
+["max_ftz_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1418,null,null,1419,1407],"4384c6cb5702956d7e64d129b6ad29ddd90e2a0909217abee23c4722bbeb1f4c",0]
+["max_ftz_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1420,null,null,1421,null],"d8d4d42ded36a100c8fda184ee7b18f999b223541b01caf20242f84053e8b0bd",0]
+["max_ftz_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1422,null,null,1423,1407],"21b20bc332c4f5651e0e22c5e946333ce0d3bbd6dd0034139e35c2910dbff415",0]
+["max_ftz_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1424,null,null,1425,null],"5163646f3af65be4ee2493ed244cb292800e8c4ac97bf2a25d929d02277c700c",0]
+["max_ftz_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1426,null,null,1427,1407],"142f1878c00309c39b91266f1d96d0e47d3000bc9797ef9ce40949e6751da200",0]
+["max_ftz_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1428,null,null,1429,null],"ae5e48a54de715f2c2e38886eb083899424ee48b5c64fed525c063f4734f44ec",0]
+["max_nan_bf16",0,0,"sm_80","+ptx70","sm_80",[1430,null,null,1431,2],"9e2e119d79b294f57f91253004d5e5dfc19415e99bd8efab9c805e9e9cfe7189",0]
+["max_nan_bf16x2",0,0,"sm_80","+ptx70","sm_80",[1432,null,null,1433,2],"2b570a76828ef09177c48c8e6d2123bf70b7fd72bc9053e807d9998f15a2b109",0]
+["max_nan_f16",0,0,"sm_80","+ptx70","sm_80",[1434,null,null,1435,2],"c56d9c0622d24c80e9620ac7ed12a89316711727ac0adc69e9f107feab80e3f0",0]
+["max_nan_f16x2",0,0,"sm_80","+ptx70","sm_80",[1436,null,null,1437,2],"067c10dd2e1e44176623f8162ed78c4e4ec2741d1989f00c3a82219e57ca8e95",0]
+["max_nan_xorsign_abs_bf16",0,0,"sm_86","+ptx72","sm_86",[1438,null,null,1439,1407],"d8950afa7ffbf2e2ca5782251be90d9229e107d84d2051b18f5ed598f175322e",0]
+["max_nan_xorsign_abs_bf16",0,0,"sm_86","+ptx78",null,[1440,null,null,1441,null],"361595cf73bfe28d7914a70e7317fd0d64037ddac2dc287368e1589ad5c7d83e",0]
+["max_nan_xorsign_abs_bf16x2",0,0,"sm_86","+ptx72","sm_86",[1442,null,null,1443,1407],"8420db4ab88ad66a5ea255b8d59185dba059b1366f078ff7aa0548016d6aba13",0]
+["max_nan_xorsign_abs_bf16x2",0,0,"sm_86","+ptx78",null,[1444,null,null,1445,null],"09317cf284b0ca87e89a04453e66df821451a8f308f3d3a709f08d0e54ed11a1",0]
+["max_nan_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1446,null,null,1447,1407],"8e768345afcee8c985fcf59a40e82826bc3356c5bd0b670b5d459fb8d187db51",0]
+["max_nan_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1448,null,null,1449,null],"0702d19b04ad538149bee81f62d662d7b203b8f88ba818d85208e5177c638d8d",0]
+["max_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1450,null,null,1451,1407],"a1d4e1e9b8b09f793e8ad16247ee34e10b65d69c8b6f50ae944debf2ff89af29",0]
+["max_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1452,null,null,1453,null],"051e557d19e25a08af965ad8e03d6aa2409e0f41fb834ca7ac6ca9335d186033",0]
+["max_nan_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1454,null,null,1455,1407],"b927bef49c2aa263ac2c28eb349f8bd0c767ae5d5fd7ab450d6f08bf0016ae19",0]
+["max_nan_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1456,null,null,1457,null],"0c4857e2b2bfb06c1cc2e95eef59084c6b58f1f500bea08973b0702b4bbcfbb6",0]
+["max_relu_s16x2",0,0,"sm_90","+ptx80",null,[1458,null,null,1459,null],"0163b41b7906af5e3e9f55ed0b78904c26d712b4df4e5c11fb49bc5d85510d49",0]
+["max_relu_s32",0,0,"sm_90","+ptx80",null,[1460,null,null,1461,null],"cd3a12164889fb25ae4340ad4a7045b9d99dc7a0f5ee72234c0ed7fc86a91fc8",0]
+["max_s16x2",0,0,"sm_90","+ptx80",null,[1462,null,null,1463,null],"e0430ddcbecb799d9552c3b6b0879891a641479829804779259ee3520192ecbd",0]
+["max_u16x2",0,0,"sm_90","+ptx80",null,[1464,null,null,1465,null],"a3558edcef85e20f0bca26f09b846d7d3aa17e7c3335ea906a04b3047adf84e8",0]
+["max_xorsign_abs_bf16",0,0,"sm_86","+ptx72","sm_86",[1466,null,null,1467,1407],"30194ecc21d74eca7bc15dfee2871110a5247385440875834d40955e28a0bcb5",0]
+["max_xorsign_abs_bf16",0,0,"sm_86","+ptx78",null,[1468,null,null,1469,null],"60d0af5fa5d16e0426f04c6e3a323d74164ec547cb9a2ddbec264b0aeaeb62c5",0]
+["max_xorsign_abs_bf16x2",0,0,"sm_86","+ptx72","sm_86",[1470,null,null,1471,1407],"e48e2f427922b4831cf91a8537a8195b32c9a48abbd8889cbde62bc9991d3029",0]
+["max_xorsign_abs_bf16x2",0,0,"sm_86","+ptx78",null,[1472,null,null,1473,null],"92d2b3546d13fbfdb0ce50ab20817947d7a2beb14667a7381484aa2d0a642064",0]
+["max_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1474,null,null,1475,1407],"723293bc036174eced8ee943ee558ca84e3cba0502dd5e16180d0f4fcbccf932",0]
+["max_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1476,null,null,1477,null],"ffc7230c08dae6edc6f3f054046a51167c0639f725450dd466d264bb89f64ac2",0]
+["max_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1478,null,null,1479,1407],"82f859dc526a37e80c1e00312a34e3e923928554c089f87ca1d8229ed0ea449b",0]
+["max_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1480,null,null,1481,null],"03caca92ec24bf5fd98dfabe1a6359658c53ec291871ae75ba412b7e325073e1",0]
+["max_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1482,null,null,1483,1407],"ae63da62b126029362cfbabf50e718d401fced2eb8a8df6aa3628ab8b8589789",0]
+["max_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1484,null,null,1485,null],"315529e2c38a950c04d22784c5693b86e7c0873aea8fbcec803914a9ec6da518",0]
+["mbarrier_arrive",0,0,"sm_80","+ptx70","sm_80",[1486,1487,1488,1489,2],"1a7db3689eb24a5f1db6b2cfac75bf993db779b94bb45ebe8181332e67f32f92",0]
+["mbarrier_arrive_cluster",0,0,"sm_90","+ptx80","sm_90",[1490,null,null,1491,14],"ccccfe3bc802b07f1cc4b3817c6ae49959601e727f0eb601b8a383f4e6dd0ee2",0]
+["mbarrier_arrive_expect_tx",0,0,"sm_90","+ptx80","sm_90",[1492,null,null,1493,14],"390a3537f9009363a38d9984d283b65e42d07964d93c980a7d17b88807447e43",0]
+["mbarrier_arrive_expect_tx_cluster",0,0,"sm_90","+ptx86","sm_90",[1494,null,null,1495,14],"81d6aae5411b1382218338673e2f636b3b9d4925ae3fa0e21f4f381cbd083a48",0]
+["mbarrier_arrive_no_complete",0,0,"sm_80","+ptx70","sm_80",[1496,1497,1498,1499,2],"f0b7267855a4e018e59fc74d9b6a1a908988b0b3ea819188f18b24a20874e9a1",0]
+["mbarrier_init",0,0,"sm_80","+ptx70","sm_80",[1500,1501,1502,1503,2],"a77acc24914f265a523cbfd8f2bb45002464f7ffcaff07598fdc73d63101a2b3",0]
+["mbarrier_inval",0,0,"sm_80","+ptx70","sm_80",[1504,1505,1506,1507,2],"31407071e2f968ef58a08b82fa472e754dfbd728d799c312bd6632cbdfac5138",0]
+["mbarrier_test_wait",0,0,"sm_80","+ptx70","sm_80",[1508,null,null,1509,2],"6c638dbd0b52d1c9ab2ce4d188aa77421bbe4c081e0a7446e4c76a91f9e59646",0]
+["mbarrier_try_wait",0,0,"sm_90","+ptx78","sm_90",[1510,null,null,1511,14],"aba700ef9d8fb5eb3dc8d85d9d440f1b8f0d72d9112df42ff3e2c30dffdfd20b",0]
+["mbarrier_try_wait_parity",0,0,"sm_90","+ptx78","sm_90",[1512,null,null,1513,14],"ee5bc939e3d59e8f515b9b60ea3890ae7d0e50027b5770e8a9897adb9beab160",0]
+["mbarrier_try_wait_parity_cluster",0,0,"sm_90","+ptx80","sm_90",[1514,null,null,1515,14],"5de038fa95cd3f60798d56ae4b5b479d004d9f35aaf3ca4d5a1ff15fed07e63f",0]
+["min_bf16",0,0,"sm_80","+ptx70","sm_80",[1516,null,null,1517,2],"96c8b2d35845a123c600893d56d86fa00bf000f976b1749fb440420597adde46",0]
+["min_bf16x2",0,0,"sm_80","+ptx70","sm_80",[1518,null,null,1519,2],"dab617569284b45df3b2e2dc1e299e9958eb7f266229c166f5ea056057118ae2",0]
+["min_f16",0,0,"sm_80","+ptx70","sm_80",[1520,null,null,1521,2],"92b2a64365779da2094972a7dd7c75d559a797e7118427a8182bf127548ba272",0]
+["min_f16x2",0,0,"sm_80","+ptx70","sm_80",[1522,null,null,1523,2],"d859f2cd15d08dc3e3fc0f7b974213c671f5217a9cd26e6e2f79dd7467980db6",0]
+["min_ftz_f16",0,0,"sm_80","+ptx70","sm_80",[1524,null,null,1525,2],"6a00adb4145e4e447413dafa01db55113db97d25c21b9ac82398620a0897954b",0]
+["min_ftz_f16x2",0,0,"sm_80","+ptx70","sm_80",[1526,null,null,1527,2],"1b025a4dd51fff82e9db6342e17c138f05c584ceee87b41780173a90cdc5467d",0]
+["min_ftz_nan_f16",0,0,"sm_80","+ptx70","sm_80",[1528,null,null,1529,2],"909d3594213b3587abc34120fffdf6efbafdd197b3119c270632efe36bbe4aed",0]
+["min_ftz_nan_f16x2",0,0,"sm_80","+ptx70","sm_80",[1530,null,null,1531,2],"f53096968c9d9c0378d5484e8040dccd28d51e0f35aab4f8c54cb31c2fa9c6f6",0]
+["min_ftz_nan_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1532,null,null,1533,1407],"8ca1bf748b08b845fc469907582b919b43e557d49fb2a3503c24f286fdf0a767",0]
+["min_ftz_nan_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1534,null,null,1535,null],"81ed98d1c0b576012d598143c879b7205af4902b77bb90a7f238ca8b580e4966",0]
+["min_ftz_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1536,null,null,1537,1407],"cc645caef42e2b3e23497b5ad5f49dd4ecc6d5a17dc9da657e94ab5a3eb6a518",0]
+["min_ftz_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1538,null,null,1539,null],"d940c5854224ac5cfc15f5dd7ae0987c384128ec877fb60d6ceab6c9d0dae913",0]
+["min_ftz_nan_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1540,null,null,1541,1407],"92e01216792eead6e05511691fddb2e5e20bf81a151e1548d5d5ed2e8ef075a6",0]
+["min_ftz_nan_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1542,null,null,1543,null],"73ea6cf485185db011f03627cb599ba4ad7e03156ac4c69d9097977e1633d805",0]
+["min_ftz_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1544,null,null,1545,1407],"d12f7e7477ff3155a5b67449389f557691f5f56ae47e4d00d7850c5da9bacc96",0]
+["min_ftz_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1546,null,null,1547,null],"3480491dc48221e52d6600eee9f14b5d0809fa6e7b2ed6ef23a1cd6748a71e54",0]
+["min_ftz_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1548,null,null,1549,1407],"2a7c4a04dfaa38fdabd20d2ca2b11237b3ba9eb8373ae5c5b5fd55774ca6befb",0]
+["min_ftz_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1550,null,null,1551,null],"c00e179a415c5fa1608db8e3eb68fa9ba401b8bb7d194e31fa3dd69017e53ccd",0]
+["min_ftz_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1552,null,null,1553,1407],"44d93aa5cb1d05acfbb247905d02683aef069b888d7f5cd3dc84820091c3f2d4",0]
+["min_ftz_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1554,null,null,1555,null],"958937784de8a545121b16c3a14d573ebc70f55331c713ee654871e7d05db339",0]
+["min_nan_bf16",0,0,"sm_80","+ptx70","sm_80",[1556,null,null,1557,2],"6b47d0215cf4b07274ec04d0bbc6ccae886d1fe477b66b9476b806e7ca32ff40",0]
+["min_nan_bf16x2",0,0,"sm_80","+ptx70","sm_80",[1558,null,null,1559,2],"9e0791b4941722b026fcf873de60cf008553990965a670e0e21b4fe7f69673b3",0]
+["min_nan_f16",0,0,"sm_80","+ptx70","sm_80",[1560,null,null,1561,2],"33e93bc2f6294ebb097449d81013445a3ac7553571d7a188596fb0923bde8f85",0]
+["min_nan_f16x2",0,0,"sm_80","+ptx70","sm_80",[1562,null,null,1563,2],"31dffc895f6161280142c5461433c8b44b1748522a9389d4c99326ac26140fd2",0]
+["min_nan_xorsign_abs_bf16",0,0,"sm_86","+ptx72","sm_86",[1564,null,null,1565,1407],"6505bd925fd8d70a9798e914fbd6ce68dc5a5eadaa75c334e1669c38c5cad7bc",0]
+["min_nan_xorsign_abs_bf16",0,0,"sm_86","+ptx78",null,[1566,null,null,1567,null],"20af51a95beafb415e8c4368324c87224f8ed5a9f7c3e5c1e4f12b62243fbee6",0]
+["min_nan_xorsign_abs_bf16x2",0,0,"sm_86","+ptx72","sm_86",[1568,null,null,1569,1407],"0e7aa15b72c08a81177eb9bfd86f90ce7e66eb603057bf91fb3b47df0b153126",0]
+["min_nan_xorsign_abs_bf16x2",0,0,"sm_86","+ptx78",null,[1570,null,null,1571,null],"6c7d624fb5a382a65d8068f7de70b08a9f1365a19f7a0edd33878fbced6eb883",0]
+["min_nan_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1572,null,null,1573,1407],"202c04f508ff7bde8896e471e9759f533af6c7f00a688727990527182d73d9d3",0]
+["min_nan_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1574,null,null,1575,null],"8b5ec83f652a77ae39fe6d8504d4ba74e1ed357147c29dee97f0a1d78f0dc026",0]
+["min_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1576,null,null,1577,1407],"34189fe11aca62a4716061076dd972afddc8524d604ea1a558d17f76e9139620",0]
+["min_nan_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1578,null,null,1579,null],"746822a130f21b753c6beb3aa123b334d98c5deaca0526e50ba1919568d0f70d",0]
+["min_nan_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1580,null,null,1581,1407],"9698bb210f04313f859020efa7c766bdcd4e350c4e01d10358433fb5137b3a5a",0]
+["min_nan_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1582,null,null,1583,null],"eddca7094a152b967c6a130d7a595b69d23b3b695b59705030fe3409031e9a84",0]
+["min_relu_s16x2",0,0,"sm_90","+ptx80",null,[1584,null,null,1585,null],"636aa6e265beccf64dbd82bb79ff9af9459fb46012f7693a2c29a3438b17831d",0]
+["min_relu_s32",0,0,"sm_90","+ptx80",null,[1586,null,null,1587,null],"5bf1da8b7efe98b99024143cd460fd56a984b828890c2b06f0258f339795611a",0]
+["min_s16x2",0,0,"sm_90","+ptx80",null,[1588,null,null,1589,null],"ab84b80cf05375a779b15d67f5b6a6c990efa79f41a96dceabac585f1af33f03",0]
+["min_u16x2",0,0,"sm_90","+ptx80",null,[1590,null,null,1591,null],"a2f7b4c40f08199470bb49b402517cd55a30b3407d3d9901dde2bbb1112dd24d",0]
+["min_xorsign_abs_bf16",0,0,"sm_86","+ptx72","sm_86",[1592,null,null,1593,1407],"c304730c79a9d11d32c8de869aceb5739ad163ef5ce27b272b062fbbf9d0f349",0]
+["min_xorsign_abs_bf16",0,0,"sm_86","+ptx78",null,[1594,null,null,1595,null],"9d20a0abfc1800894c8cb613ec0e8f33285bc0e49e3776617f44c4acba97e673",0]
+["min_xorsign_abs_bf16x2",0,0,"sm_86","+ptx72","sm_86",[1596,null,null,1597,1407],"fa7646c35cd0560c2728dac1b04d3a5cc216fe0356c00e702f24747d568934c0",0]
+["min_xorsign_abs_bf16x2",0,0,"sm_86","+ptx78",null,[1598,null,null,1599,null],"84004b592554504f3d84a09bbf05ae14df4373063f2d6535df0ea97b162acb66",0]
+["min_xorsign_abs_f16",0,0,"sm_86","+ptx72","sm_86",[1600,null,null,1601,1407],"8197512d5c316d90e590da10a1f8f88c59fe20348353f1e56da98dc2bb0f9a28",0]
+["min_xorsign_abs_f16",0,0,"sm_86","+ptx78",null,[1602,null,null,1603,null],"c2092055af54c40177960ce3edb1b9dc1cb59afdcda717e464bc7233c1203daa",0]
+["min_xorsign_abs_f16x2",0,0,"sm_86","+ptx72","sm_86",[1604,null,null,1605,1407],"81f72dfc6391c13cd5b44dbf3b1a03d16e29b0c61f0fc468b07ef7ffb3d43f3b",0]
+["min_xorsign_abs_f16x2",0,0,"sm_86","+ptx78",null,[1606,null,null,1607,null],"ba6390531e2965acae8dd9577c705a3d4308b4e2f0db9e92fb776a4040621b2f",0]
+["min_xorsign_abs_f32",0,0,"sm_86","+ptx72","sm_86",[1608,null,null,1609,1407],"25665cee4fd5afe966830a7e28b5906dec1de0aa5d8c9c4575a261b3c664b390",0]
+["min_xorsign_abs_f32",0,0,"sm_86","+ptx78",null,[1610,null,null,1611,null],"7a1361b76fa662b22d301814cd6d347c008c24fc107c45208bb609262c874333",0]
+["mma_m16n8k128_s32_b1_and_popc",0,0,"sm_80","+ptx71","sm_80",[1612,null,null,1613,2],"13b634dcd229f5014f137ddda3a6299efdc727dc529c4344732a7fb889efe2b3",0]
+["mma_m16n8k128_s32_b1_xor_popc",0,0,"sm_80","+ptx70","sm_80",[1614,null,null,1615,2],"80f6321e2889f064b1020987922a35860731bb2027ce0f301f053253f7889d33",0]
+["mma_m16n8k16_f16_f16",0,0,"sm_80","+ptx70","sm_80",[1616,null,null,1617,2],"e8bd19aa4edac454dff23a0b3eb7a5ba4b4957d19b1397540b1bd4c4738a1254",0]
+["mma_m16n8k16_f32_bf16",0,0,"sm_80","+ptx70","sm_80",[1618,null,null,1619,2],"7a2e20022ab13a6388ba5f038cd1e98914faa3fa84607bbcc37e9f5c43fda28d",0]
+["mma_m16n8k16_f32_f16",0,0,"sm_80","+ptx70","sm_80",[1620,null,null,1621,2],"53d073813eec58778d63cd17721408cf14a48dbc8f72e70f7e3edd78237adcc8",0]
+["mma_m16n8k16_fp8_f16_e4m3_e4m3",0,0,"sm_89","+ptx87","sm_89",[1622,null,null,1623,714],"4f41a17049d1e7563ee849c65dd4f6d223dc4e656c7ff565ac4f3598434ee364",0]
+["mma_m16n8k16_fp8_f16_e4m3_e5m2",0,0,"sm_89","+ptx87","sm_89",[1624,null,null,1625,714],"0b68cee2ade432c2a730724eb00fb8442ebedf63d1aa731428e04bf554d89830",0]
+["mma_m16n8k16_fp8_f16_e5m2_e4m3",0,0,"sm_89","+ptx87","sm_89",[1626,null,null,1627,714],"d80c8162f86493db4ecf79cf785db0c3f1e0766a85a90ad2d52ed0f83490dba0",0]
+["mma_m16n8k16_fp8_f16_e5m2_e5m2",0,0,"sm_89","+ptx87","sm_89",[1628,null,null,1629,714],"3901c8997d06eb7ea64c2ef7811e4e585afa3b064b14e8c70aa725bab1840955",0]
+["mma_m16n8k16_fp8_f32_e4m3_e4m3",0,0,"sm_89","+ptx87","sm_89",[1630,null,null,1631,714],"a4c2fbbc5190861c3a1adf287082dbae27d359ebd55eae9b61fa839c15dac95a",0]
+["mma_m16n8k16_fp8_f32_e4m3_e5m2",0,0,"sm_89","+ptx87","sm_89",[1632,null,null,1633,714],"9e734f6fd1b7dcb4aa746f63131ce679118ec421877b5e303365c3da4deb5530",0]
+["mma_m16n8k16_fp8_f32_e5m2_e4m3",0,0,"sm_89","+ptx87","sm_89",[1634,null,null,1635,714],"8f5cf2e751117e2c22cdcd7ad4a9efed62e54605977d0d2844c7a9ae408e8c98",0]
+["mma_m16n8k16_fp8_f32_e5m2_e5m2",0,0,"sm_89","+ptx87","sm_89",[1636,null,null,1637,714],"b13a3e148ec8f03a8c7e04c882072ddebcb37ae11676550e293f345087a116f5",0]
+["mma_m16n8k16_s32_s8",0,0,"sm_80","+ptx70","sm_80",[1638,null,null,1639,2],"0a17a593ca54ade345026e77626ee688dcdae36699c2f8839caf777c810ce7a2",0]
+["mma_m16n8k16_s32_s8_satfinite",0,0,"sm_80","+ptx70","sm_80",[1640,null,null,1641,2],"5022ccedd3211065235dad03cfc97569d57d61cbc6927967cb41ed95a3b41b5d",0]
+["mma_m16n8k16_s32_s8_u8",0,0,"sm_80","+ptx70","sm_80",[1642,null,null,1643,2],"6897a7f3e7bff9aacd9493bff395b1cc2f905ee05e3386a071abd9f6fc854f87",0]
+["mma_m16n8k16_s32_s8_u8_satfinite",0,0,"sm_80","+ptx70","sm_80",[1644,null,null,1645,2],"86b39b4f881f6c1b276eaeaf4c9018f6dd66cdbb134126b29e2c4e023fa4b652",0]
+["mma_m16n8k16_s32_u8",0,0,"sm_80","+ptx70","sm_80",[1646,null,null,1647,2],"e1a62ea36433b0e2b9638fc6f66767f7895038aea47ebcb172ec2474d71e000e",0]
+["mma_m16n8k16_s32_u8_s8",0,0,"sm_80","+ptx70","sm_80",[1648,null,null,1649,2],"fc877b87e37ab9bc329283627253ef6edd2876317df1257a8fbd25fbf5d59c6e",0]
+["mma_m16n8k16_s32_u8_s8_satfinite",0,0,"sm_80","+ptx70","sm_80",[1650,null,null,1651,2],"401e538646c99198f18175becc0fec1113d826526d91f424ab832f2d44edd9c7",0]
+["mma_m16n8k16_s32_u8_satfinite",0,0,"sm_80","+ptx70","sm_80",[1652,null,null,1653,2],"a8c4f7a2697edd1556c5e0cc4555b883865928b70f62d17022d2a8412440e684",0]
+["mma_m16n8k256_s32_b1_and_popc",0,0,"sm_80","+ptx71","sm_80",[1654,null,null,1655,2],"9034c74dd081ac02b13014bfe4d67600aa28a11d2165f6a41eb20303b6e4a89f",0]
+["mma_m16n8k256_s32_b1_xor_popc",0,0,"sm_80","+ptx70","sm_80",[1656,null,null,1657,2],"2affb83f2abeecaea80f650c4ed6283b56826072ea89fafe40235ee9b02a9562",0]
+["mma_m16n8k32_f16_e2m1_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1658,null,null,1659,1091],"f1574d93067201485809e9ec606dc2d6204b0f4d94b61a652e2ae67156707147",0]
+["mma_m16n8k32_f16_e2m1_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1660,null,null,1661,1094],"8cca4453f72a40845a63febd17477533d1bb076970aaa58ebfdcece907f45a6a",0]
+["mma_m16n8k32_f16_e2m1_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1662,null,null,1663,1097],"22027300afbdf334caf59063b530c9a1186d9ec4439e1c408c4b27c0e5d5d9ab",0]
+["mma_m16n8k32_f16_e2m1_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1664,null,null,1665,1100],"71bfc275b753adbb4ea3d1008a5756e62a8d51e9f33d6fcfcb07903963524b0d",0]
+["mma_m16n8k32_f16_e2m1_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1666,null,null,1667,1091],"4dd20fe694d17897e4f77c7cd73f5c29899405181271e73f2df910e137d78878",0]
+["mma_m16n8k32_f16_e2m1_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1668,null,null,1669,1094],"9a0d31a786bbb80aeb258f00bd8166329543167bf1716b332f45b23c325ef319",0]
+["mma_m16n8k32_f16_e2m1_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1670,null,null,1671,1097],"8bb02824990cf9044828d7583813c0ae1f93bf2435fe868f68a448ea28edc9b1",0]
+["mma_m16n8k32_f16_e2m1_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1672,null,null,1673,1100],"edef2a77d404a2b4139c23210c89183819945b58ec7f99f6f36a33797cadfc11",0]
+["mma_m16n8k32_f16_e2m1_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1674,null,null,1675,1091],"8efd1a4d7e09fd0b9e67caf76b884900a1e729e8af64818025eb20ba78fdd008",0]
+["mma_m16n8k32_f16_e2m1_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1676,null,null,1677,1094],"8646cccb32e01c11f61fa3edd0c3913ee5e7e0c8a5fb99ed8312fe6ca2f90934",0]
+["mma_m16n8k32_f16_e2m1_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1678,null,null,1679,1097],"5f508a97a9f69262c76e24c48ee94779998a82b4d40ec3b2ea6e6be36879b821",0]
+["mma_m16n8k32_f16_e2m1_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1680,null,null,1681,1100],"16aa877692e580d3c243d4e8ed5b0213cc9fed24cd1738be74b5d3f1f007ef85",0]
+["mma_m16n8k32_f16_e2m1_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1682,null,null,1683,1091],"bedabab22f8c5893539eed0cfa8635ccd1a7473c16772ef9c93b39f988080023",0]
+["mma_m16n8k32_f16_e2m1_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1684,null,null,1685,1094],"d8b3c821d6d6096ed9590d8b8f02c54402845dba7ca8ab72823a0a2bf0da3b9d",0]
+["mma_m16n8k32_f16_e2m1_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1686,null,null,1687,1097],"05680bc3b786f8e5e26e8a43ba4cead81540e9e6fc8d9fbf6c106024fbce22d8",0]
+["mma_m16n8k32_f16_e2m1_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1688,null,null,1689,1100],"6f6d5d89d94a019878ed1c77b68df548d289f57bb4d1d78f252dcb524ff4469c",0]
+["mma_m16n8k32_f16_e2m1_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1690,null,null,1691,1091],"b092aa92734d0d26ee08c01a9f5b7ab9843bf246af90f119d7c9fad61b000a77",0]
+["mma_m16n8k32_f16_e2m1_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1692,null,null,1693,1094],"3bb3cb0bfe1cc0b03fd8d837b64e28cd20ab873e8c0bb0457fd760c26ede4625",0]
+["mma_m16n8k32_f16_e2m1_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1694,null,null,1695,1097],"982ccb48d319afae55e8d58ad131c3aa5c732224f8e57fc53e541a0b17678298",0]
+["mma_m16n8k32_f16_e2m1_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1696,null,null,1697,1100],"5fbfd84a0dfa16f6f24e3566f668796d6f314c6c60a26cd33410b54aa665a255",0]
+["mma_m16n8k32_f16_e2m3_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1698,null,null,1699,1091],"540a206f7b2dd948370a187413a941a5549d2a1072332bba878861c40f191458",0]
+["mma_m16n8k32_f16_e2m3_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1700,null,null,1701,1094],"a8879e06a21b64174849904d0e5e289debc5e2f65ef7baf4d34bae96370b367b",0]
+["mma_m16n8k32_f16_e2m3_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1702,null,null,1703,1097],"00956a55aee4c9ba033f84f6d3d8f84b145fd953ef6ff1c18b35b69d9b914a8b",0]
+["mma_m16n8k32_f16_e2m3_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1704,null,null,1705,1100],"d7287371338c8cf6dfd4f89cc9287ab5caa63a758584db6361fe026eacda6295",0]
+["mma_m16n8k32_f16_e2m3_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1706,null,null,1707,1091],"880507dfc2ab287539839e46638cc5508a7a6c8240ede800487cfb19057aef35",0]
+["mma_m16n8k32_f16_e2m3_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1708,null,null,1709,1094],"446cc56c8a9f7fc79d930510c2a164112000495f0820d868ce7afa9482f1d1ea",0]
+["mma_m16n8k32_f16_e2m3_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1710,null,null,1711,1097],"7f8e8eff5f50718f477e0ba97ffbc1fde1ec80ab0ce84e8b8e5a28d04dd6b3f6",0]
+["mma_m16n8k32_f16_e2m3_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1712,null,null,1713,1100],"340a0ae78e77f6c653ec4e9073bdb044225ec03e7301ef4075e27e584ecb0b98",0]
+["mma_m16n8k32_f16_e2m3_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1714,null,null,1715,1091],"d301817185ea3f099ea3a627a6f62fa79fd5efdae6d85ae45abbb9837f65e99b",0]
+["mma_m16n8k32_f16_e2m3_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1716,null,null,1717,1094],"aed793f50b534139a542f6279b8b4ea0d9fa242cad8952499854250b08340ac6",0]
+["mma_m16n8k32_f16_e2m3_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1718,null,null,1719,1097],"a6bbabc4b64a1017d9ef6d4d0decc6c958c482fa77610f2d7b6a4b9c1e769b0d",0]
+["mma_m16n8k32_f16_e2m3_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1720,null,null,1721,1100],"b9c7341769e3bea2e8d548b57675e5940948d9614ce4eda82167d3dd56388f58",0]
+["mma_m16n8k32_f16_e2m3_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1722,null,null,1723,1091],"8795db230ba15abf04fc6cef935bfa0b11fe33b81fe99151d69a1e377ac691b7",0]
+["mma_m16n8k32_f16_e2m3_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1724,null,null,1725,1094],"eed5b7e97d4480675b56587f17678484487ecd8f5fbebd393fb62a796b8fa2bc",0]
+["mma_m16n8k32_f16_e2m3_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1726,null,null,1727,1097],"e797edcb0362498ba06a6f1502eb9401a3ca4524049e7a23167ad9c769b5fb13",0]
+["mma_m16n8k32_f16_e2m3_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1728,null,null,1729,1100],"959cad0aed5cdfe694ddc80a0983367916d8ec5aff02fc8a38db1900af8a3843",0]
+["mma_m16n8k32_f16_e2m3_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1730,null,null,1731,1091],"99b520e24c01172e28bfcae2f5b9e0a3e1fd2038c5e5ad2d18cd49f218bf2e1e",0]
+["mma_m16n8k32_f16_e2m3_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1732,null,null,1733,1094],"d9cccf282422031e41b76c2f50ea9044117faba3f1ef3dc8894428fc9eb87652",0]
+["mma_m16n8k32_f16_e2m3_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1734,null,null,1735,1097],"06b17810a7c62c34f16c565110cd3bcfeb69b8935abdfec76fab5203c3f6c7cc",0]
+["mma_m16n8k32_f16_e2m3_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1736,null,null,1737,1100],"57c7e2cc082b37224225891e4a8d0429a0519b4b6d94e61c5635a222acf26002",0]
+["mma_m16n8k32_f16_e3m2_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1738,null,null,1739,1091],"50796affb978682e31ea245a344bb8de3c8d9eb6b2249835a9b683c953e1d580",0]
+["mma_m16n8k32_f16_e3m2_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1740,null,null,1741,1094],"3fc33286ad7403ae3b0d093e69fa7771d6aa1a4caf1da3f790a9362f3272a5d5",0]
+["mma_m16n8k32_f16_e3m2_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1742,null,null,1743,1097],"9058a7ef289bc4a4701af7ca1c8d05970b3673c624179381511048e592546dc3",0]
+["mma_m16n8k32_f16_e3m2_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1744,null,null,1745,1100],"5049f8620f6a5e26d5ffa96d719a429921d2bdb3f190abb98fc26bbac427c71b",0]
+["mma_m16n8k32_f16_e3m2_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1746,null,null,1747,1091],"d9d97c51641d122a4a41558a5a3cac9f14bb6c7d78dcb0ca88f301284938dea1",0]
+["mma_m16n8k32_f16_e3m2_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1748,null,null,1749,1094],"1172a52a84d8292db6c7992955799db2df75e1708fca24b90afbffbeb1795961",0]
+["mma_m16n8k32_f16_e3m2_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1750,null,null,1751,1097],"1496fd05da3af74300faf963f28359bc891eac0d123d802f76caf255153c572f",0]
+["mma_m16n8k32_f16_e3m2_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1752,null,null,1753,1100],"f43c9ee861e013a2c1515c58bd085fc17124350b30ddefd65914f62dd22b6b68",0]
+["mma_m16n8k32_f16_e3m2_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1754,null,null,1755,1091],"4e9dad78130c7525f88a3a7e7c9eda2b6d6e9e76197e2e35bfa1d3df3c4ce12d",0]
+["mma_m16n8k32_f16_e3m2_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1756,null,null,1757,1094],"3302891d53c18541ca6b7f8149d10b5c51109880914c4a2ab50f545223d92e62",0]
+["mma_m16n8k32_f16_e3m2_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1758,null,null,1759,1097],"b07c27a5557452c9b4e7c24872c4a99252ea424de0c8dfc759b9e59c7b654397",0]
+["mma_m16n8k32_f16_e3m2_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1760,null,null,1761,1100],"64a5986649111ca24bc8e5835c257333553ad2356f297b2d8fde329a4108a7bd",0]
+["mma_m16n8k32_f16_e3m2_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1762,null,null,1763,1091],"ee2c06b0b6dadcc39b49117334d00556c8008cf2101e381c75b5f5a0d9efe3e5",0]
+["mma_m16n8k32_f16_e3m2_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1764,null,null,1765,1094],"926214b5cd3a28f470ac45b0200c57dd67b817bc21908070c2132ac08a240ffd",0]
+["mma_m16n8k32_f16_e3m2_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1766,null,null,1767,1097],"6d15ea349ef0c20393f25f365ede09e72129eafd5b4d28fca600e90808090213",0]
+["mma_m16n8k32_f16_e3m2_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1768,null,null,1769,1100],"4460fa8f7bec5419a978c1c4d0ac5793505cdf29be276d9ee1484a634b8cb86f",0]
+["mma_m16n8k32_f16_e3m2_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1770,null,null,1771,1091],"c7e03a5adac4e02aac93c4e5a0a846aa0f9e018843291824380c39be2bd1a251",0]
+["mma_m16n8k32_f16_e3m2_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1772,null,null,1773,1094],"a742858da8be38a7242607bd325b16fd3577af49addfe0c421ca2f0f7be323bc",0]
+["mma_m16n8k32_f16_e3m2_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1774,null,null,1775,1097],"8002eb20e800eb809d4861aa03ad4e7a79b9145d8f24ae1ec93b90b3f8db581a",0]
+["mma_m16n8k32_f16_e3m2_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1776,null,null,1777,1100],"622bc9b61529b3bf4df1721de0b0e0bc48b35742689e75ccbdb133f320e8ef62",0]
+["mma_m16n8k32_f16_e4m3_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1778,null,null,1779,1091],"e77719916bfae85a60cdb850847cac66ebdb6d0b919ff3d2a47aee776f9388d6",0]
+["mma_m16n8k32_f16_e4m3_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1780,null,null,1781,1094],"79781656809e6b305d2606985ba5e5ded703523fd1925e608c4812bffbc935f4",0]
+["mma_m16n8k32_f16_e4m3_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1782,null,null,1783,1097],"cfa54550b34a9cf745040b68f305a4f022c1d7df3816bbe1dab88bd442e10380",0]
+["mma_m16n8k32_f16_e4m3_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1784,null,null,1785,1100],"a8d25d73f69f80db06e7965ab76b2b4659332a089608497d49ef620e5ba1c447",0]
+["mma_m16n8k32_f16_e4m3_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1786,null,null,1787,1091],"e11e39d8a3df233ed0de120e8b8ba337f49d026cc2fe13b1d95bc3522cc446dd",0]
+["mma_m16n8k32_f16_e4m3_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1788,null,null,1789,1094],"98c39d47a14a1c637df1f898a85e061f7715ff976b25f0456e818e1666f43431",0]
+["mma_m16n8k32_f16_e4m3_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1790,null,null,1791,1097],"58b34524fda12685855235bb6d802117a713d864ecbaf6b16eb12884a65a3455",0]
+["mma_m16n8k32_f16_e4m3_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1792,null,null,1793,1100],"bdf065a8e847a1c0614b9f0ba32fbf360231015dde1b4e2a79c74f3a7b5c633c",0]
+["mma_m16n8k32_f16_e4m3_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1794,null,null,1795,1091],"616916484b560a591abbe1bed78cbf7bb0d79cf30104bc4c0dd280474735699a",0]
+["mma_m16n8k32_f16_e4m3_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1796,null,null,1797,1094],"fd9b3b85fe094a9871a377a29a6c475f73e28938a989c125c7ec992546865717",0]
+["mma_m16n8k32_f16_e4m3_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1798,null,null,1799,1097],"bc16dc7a72328e0f9be936b3618aa69810c6730315e097f6b53c7e158b317fe1",0]
+["mma_m16n8k32_f16_e4m3_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1800,null,null,1801,1100],"0f69babc726ca743f37ebcd030cfe9c89315a54617b5ba2d9f6bba485429ea44",0]
+["mma_m16n8k32_f16_e4m3_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1802,null,null,1803,1091],"2f93f11d92ebad5c7e7ae03c7eb74b5548d8a1c7d3bc5887493e374f1b21bd7e",0]
+["mma_m16n8k32_f16_e4m3_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1804,null,null,1805,1094],"83451a3f67e8351dbfc3ddae5d215a54910136ead6311c01b0bb27420cf128e2",0]
+["mma_m16n8k32_f16_e4m3_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1806,null,null,1807,1097],"8ef63fc70d3a577a3a65f30fac6dc85698f0ae4c8d5400a382bc341ed20a4c6d",0]
+["mma_m16n8k32_f16_e4m3_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1808,null,null,1809,1100],"501639621486ca92687eb479b991e9f83029bbe073d7128691ac614fc5fcd01a",0]
+["mma_m16n8k32_f16_e4m3_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1810,null,null,1811,1091],"6cf742818373bb9bd62a641b0833d50fd85e8f7c91936850748bf1dc014b0e5f",0]
+["mma_m16n8k32_f16_e4m3_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1812,null,null,1813,1094],"310b7fa118eff1c9b10711220313f052fbdf0c26f3eaefcbe85358f50bc29224",0]
+["mma_m16n8k32_f16_e4m3_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1814,null,null,1815,1097],"d6d88aeb61a955f696e4a6186ed70ce8322f53aa9e20a0d66da222996d297193",0]
+["mma_m16n8k32_f16_e4m3_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1816,null,null,1817,1100],"d69b6caabc777e91cc71141b3f30c0604a49690313854482184cdbe3689101ea",0]
+["mma_m16n8k32_f16_e5m2_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1818,null,null,1819,1091],"567095198cbd3fb80287a29e0422c1a59bc6cca03749d568128e6bc136aff64d",0]
+["mma_m16n8k32_f16_e5m2_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1820,null,null,1821,1094],"2f3c2d3f0a514140d3d5013530c0b1d76a2f454ef8be94d0a02955ab981c73e5",0]
+["mma_m16n8k32_f16_e5m2_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1822,null,null,1823,1097],"435ca4c7964d6faff47963515a1d74cf9e9c92ef885217371cf43641f61b018f",0]
+["mma_m16n8k32_f16_e5m2_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1824,null,null,1825,1100],"c64c06619911a748babd57e8143e0d6013408664aa7425fa516731f71546f3fa",0]
+["mma_m16n8k32_f16_e5m2_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1826,null,null,1827,1091],"89ef36e8bb79ec035f1d98e265f2bc94c2c1b1c1721d5d70a27ee3e897106777",0]
+["mma_m16n8k32_f16_e5m2_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1828,null,null,1829,1094],"58cd8f942e8c747892a881c0affc3cc4a1dfc1766e20cd0cc9956972db562895",0]
+["mma_m16n8k32_f16_e5m2_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1830,null,null,1831,1097],"5e245c7fb03d68946fc737993ba2b48077485ab1546e274751de865702abe1d5",0]
+["mma_m16n8k32_f16_e5m2_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1832,null,null,1833,1100],"acce8f523710f8b5c619455c14324360f08afe01c0453675389a124d961d551e",0]
+["mma_m16n8k32_f16_e5m2_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1834,null,null,1835,1091],"fbe0940da3c146ad41a7e037d254d2e0c88600c8b9d6ff56f90aa22a4f8d8aca",0]
+["mma_m16n8k32_f16_e5m2_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1836,null,null,1837,1094],"76b6bb97af9378f5aed671152c9ac1fc493687d64d68297f450045f418805211",0]
+["mma_m16n8k32_f16_e5m2_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1838,null,null,1839,1097],"4e1e0d89c6651cbc92e1694ed87ea68823f41ae7dd745e6bd2a3978fb1e35502",0]
+["mma_m16n8k32_f16_e5m2_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1840,null,null,1841,1100],"91b8cf96313ddbd02d921a3127a8f59be21756fd3884b811956054b58fd01bcc",0]
+["mma_m16n8k32_f16_e5m2_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1842,null,null,1843,1091],"ba790500d214dad5a5ef84e7af432242da3937d27b50df7b322eb83764e89ff3",0]
+["mma_m16n8k32_f16_e5m2_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1844,null,null,1845,1094],"863e09e349a6a6f9371a469a0f6b2e41ebe0d9cfb6eceff493b8faeb012ceac5",0]
+["mma_m16n8k32_f16_e5m2_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1846,null,null,1847,1097],"0116ea030eecaa223d2be18e0595c28b39123895f6c68418da093846a09dc873",0]
+["mma_m16n8k32_f16_e5m2_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1848,null,null,1849,1100],"462e0306213b21384f56fb9c0775fd9b2392f6008831a2762c58b451ad2faba5",0]
+["mma_m16n8k32_f16_e5m2_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1850,null,null,1851,1091],"8d4c8ff6ae9131890379974664d8f39789f948854a04fa8106ff10a60cf0b736",0]
+["mma_m16n8k32_f16_e5m2_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1852,null,null,1853,1094],"481b4c8cadd352598ac497ede7e97094c30dd21adcd3775d391149ce548bcb04",0]
+["mma_m16n8k32_f16_e5m2_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1854,null,null,1855,1097],"5777ecd42c3d7d4a5f5ee28c2774789cc51a4115adeabe4d7cdce8834a258b34",0]
+["mma_m16n8k32_f16_e5m2_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1856,null,null,1857,1100],"966b04e1bda142c67ca2cf5b058e0e78d9492e0891d798a94a7b6956af2c4f3a",0]
+["mma_m16n8k32_f32_e2m1_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1858,null,null,1859,1091],"98eb4bd61264db931016fd7541212b8f37608dafc6f4b7bb8188b8aaed5919b1",0]
+["mma_m16n8k32_f32_e2m1_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1860,null,null,1861,1094],"ebd6e916554b2e690b8f7517a35ba14a1544f0cf3d6130ce1d8c81957145f69f",0]
+["mma_m16n8k32_f32_e2m1_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1862,null,null,1863,1097],"d0a0242af5e4a6684887baedbb4515f8335a2babd0ea37e7a5103b2bf051c7ab",0]
+["mma_m16n8k32_f32_e2m1_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1864,null,null,1865,1100],"12de35b92f89433e5a1d5cebd717e6099c47722ee51a3b893ef00c5233e3dfb2",0]
+["mma_m16n8k32_f32_e2m1_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1866,null,null,1867,1091],"2116ab07f82d6736f740bbfee6fc62962fb7a1288c74118efb5f0459f6057590",0]
+["mma_m16n8k32_f32_e2m1_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1868,null,null,1869,1094],"41eb7ee31e3a9468a6a4e9b7624423111d9b2a6fcac866d799a3111441593131",0]
+["mma_m16n8k32_f32_e2m1_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1870,null,null,1871,1097],"38578cfb2c6cc2b143db9577b01e1572f0266f9ae91aa07cbac3e1480802e1b6",0]
+["mma_m16n8k32_f32_e2m1_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1872,null,null,1873,1100],"11cb60acb3691b6d8187c4176a512a726f1c3ec03b567fd6d08a6d0dccadc772",0]
+["mma_m16n8k32_f32_e2m1_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1874,null,null,1875,1091],"91852b6db1d3921f3822145bd579b1ab04af67d052bedce6817876341f87106c",0]
+["mma_m16n8k32_f32_e2m1_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1876,null,null,1877,1094],"0f1325981833a9d63124d34263833e895af41e93da314789b492c172a518b6f8",0]
+["mma_m16n8k32_f32_e2m1_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1878,null,null,1879,1097],"69e67aa9a913ef61428ab4dd633fdca43b111b73eae2d571bffbdecb3a105e8f",0]
+["mma_m16n8k32_f32_e2m1_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1880,null,null,1881,1100],"d66807d2079407ae19efec567bd47ad55047a611318bfe59c2862293eae07a9c",0]
+["mma_m16n8k32_f32_e2m1_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1882,null,null,1883,1091],"e1f0d74fca486f8c4c82cfa9f63ca7d24c2bcfbc48f2239b6ca21e4dc0598b99",0]
+["mma_m16n8k32_f32_e2m1_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1884,null,null,1885,1094],"ca5654adc8fabbe3afcd172b7d9747e7de34a99747e2456fe0b332f9851949f6",0]
+["mma_m16n8k32_f32_e2m1_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1886,null,null,1887,1097],"8548e6c550c38d84ebd0cd77ba916f410200061a2e68be1d2950ec35f1502abb",0]
+["mma_m16n8k32_f32_e2m1_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1888,null,null,1889,1100],"5ace9df4997728d56e4e5c90cd57b23a22d5aa2f4bff073dfbd0abb09ec8cb94",0]
+["mma_m16n8k32_f32_e2m1_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1890,null,null,1891,1091],"8bb467ade236615171643b7b0858aa8e60580d88da550a7e1fd2a601637d9cf7",0]
+["mma_m16n8k32_f32_e2m1_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1892,null,null,1893,1094],"1789f205fd4f4750949187e58a0b0196f3f74b1f321152de1ed22d802491e344",0]
+["mma_m16n8k32_f32_e2m1_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1894,null,null,1895,1097],"8bb8d044faa02776e5cd75002ea697558efa299d9272c66576efe8bea985a62b",0]
+["mma_m16n8k32_f32_e2m1_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1896,null,null,1897,1100],"a1e8c5f455ed8eb1e13ee482666efaa0224d7805b20de395b9d385b7cccb7735",0]
+["mma_m16n8k32_f32_e2m3_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1898,null,null,1899,1091],"e76e71e0eb14daa4c9e1a22f867d700473511c8accebe20c3929562f33a5fa29",0]
+["mma_m16n8k32_f32_e2m3_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1900,null,null,1901,1094],"244329209684bc9daed93de7d449b9ccb0b9449b3102868e642686630bdef940",0]
+["mma_m16n8k32_f32_e2m3_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1902,null,null,1903,1097],"adc95a487fc7c0a20ad2ad792fbc96c55504c33ddc9d37618f0d2d99a17644bb",0]
+["mma_m16n8k32_f32_e2m3_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1904,null,null,1905,1100],"52443d5e3f5dba394ee46842175deb143646622a528b20d3d1c89e2bc852ff11",0]
+["mma_m16n8k32_f32_e2m3_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1906,null,null,1907,1091],"acb3be08623f575cf6ba93db230bb74d7d4ede6b1c922fac49b86f1cb117cc59",0]
+["mma_m16n8k32_f32_e2m3_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1908,null,null,1909,1094],"415b8a886fab886b50f5371511f0d92a30d44bb4fc93496d8a6d84fef1fa9df6",0]
+["mma_m16n8k32_f32_e2m3_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1910,null,null,1911,1097],"1dec6f85f50c82f094c1e5ebab9f0a98da91d7baa1021f942668b1ab4283618b",0]
+["mma_m16n8k32_f32_e2m3_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1912,null,null,1913,1100],"b36e03047bec87c6dd9fddc81c597aa7d2b3de0f83b75b448451c1f9187ab270",0]
+["mma_m16n8k32_f32_e2m3_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1914,null,null,1915,1091],"bd2880f7a15734f1ee2047cbb179a0165da363f26c82a357a8aaf00c3ecb463f",0]
+["mma_m16n8k32_f32_e2m3_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1916,null,null,1917,1094],"d8e2e16d379f2e2f08ed5d7e89e70a15e20fd69175321bb399c399cf34a3e9d6",0]
+["mma_m16n8k32_f32_e2m3_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1918,null,null,1919,1097],"36d8373574faa3cf74ab2c84668c54b5186a42ec7d3e37056065c7d73edab4fa",0]
+["mma_m16n8k32_f32_e2m3_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1920,null,null,1921,1100],"c5fcd56311b74e56f5b75dfd9ea68bfa2bc05a1605609eff798ed385e872831d",0]
+["mma_m16n8k32_f32_e2m3_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1922,null,null,1923,1091],"61781a4c6ff8392d9f45c8e1b7b80e655cc57197888cf4f1527563bde5dc1df8",0]
+["mma_m16n8k32_f32_e2m3_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1924,null,null,1925,1094],"3651b978c92542fd04ea51fed2d2f5456e7c103739e6c3ecbb9ec8173ff9d96f",0]
+["mma_m16n8k32_f32_e2m3_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1926,null,null,1927,1097],"449cb32187ee978535796c5dee470055fb2fbcebfc929c28b1e242793865a333",0]
+["mma_m16n8k32_f32_e2m3_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1928,null,null,1929,1100],"0fd955022afcb3cb2365db844b3eb44f9f3ef0e622c0706da038df8871a77f40",0]
+["mma_m16n8k32_f32_e2m3_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1930,null,null,1931,1091],"8d8fefd18e4d9513b3fc801b2ad491e4b4e10106f4cb0e2b0f6a481aa6538bff",0]
+["mma_m16n8k32_f32_e2m3_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1932,null,null,1933,1094],"62bd36b8385d0c326776522650fbc1e09769332e1243c6f15e6f260a8d9fa786",0]
+["mma_m16n8k32_f32_e2m3_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1934,null,null,1935,1097],"a913fb97a61653dcecace27d6d37970df8d0411a7492cbaf0c981ccefb6e172b",0]
+["mma_m16n8k32_f32_e2m3_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1936,null,null,1937,1100],"d276d8f7b1886707c6e7730741bc24d9c8e6e94db8907442922f8e9527e43220",0]
+["mma_m16n8k32_f32_e3m2_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1938,null,null,1939,1091],"55d68d962d0292d754dfdfb2a6028dc7ec682e8e283f8df7db0a25be24bbdf32",0]
+["mma_m16n8k32_f32_e3m2_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1940,null,null,1941,1094],"aadd81266fb729b232e93a02639cde7ab9deed9f66a4cbaeb6624f1adfcac739",0]
+["mma_m16n8k32_f32_e3m2_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1942,null,null,1943,1097],"1fd86ea1cb1dbaf6d7777b7c6facce5ba95c52309b4b6831036deb0604fd312d",0]
+["mma_m16n8k32_f32_e3m2_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1944,null,null,1945,1100],"4e7cb81b44fbf6b20364b0be4847430ccee030dbeb46f9dd522a480833436342",0]
+["mma_m16n8k32_f32_e3m2_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1946,null,null,1947,1091],"06b4aa6156d585cfe812c0cb8e85eebc57bd0cb015fcd8123dc040672d05c4ff",0]
+["mma_m16n8k32_f32_e3m2_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1948,null,null,1949,1094],"7d3993a2b0beca6a2acf94c0f5eb6cc888a82738739f2a348ee37117c849d10f",0]
+["mma_m16n8k32_f32_e3m2_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1950,null,null,1951,1097],"b4eff1e515f4f1da6b3230f763adb37a0d0de95f56a39503051395c1e94fb4c4",0]
+["mma_m16n8k32_f32_e3m2_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1952,null,null,1953,1100],"7288a619dd2e816c0ac17dccdd42e8f991555598d7d532add5cdacfc5614d261",0]
+["mma_m16n8k32_f32_e3m2_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1954,null,null,1955,1091],"f9a62c1ea81643eaef748b8b42dd38de9f34b980578a14f653ad6964e84d584d",0]
+["mma_m16n8k32_f32_e3m2_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1956,null,null,1957,1094],"e2a08da31d4bdd9c1eff9d758e8c1fcbadb44b52739f305db90ba8b2a9320176",0]
+["mma_m16n8k32_f32_e3m2_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1958,null,null,1959,1097],"78b0039cd45f5716e6ab5a6dbf78377ab8ec6a2de6200c899925b7d65362b867",0]
+["mma_m16n8k32_f32_e3m2_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[1960,null,null,1961,1100],"1951e810a9eb95e3f2a7cd6a054e0929f96f5e5427aa9ad543834641f4735457",0]
+["mma_m16n8k32_f32_e3m2_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[1962,null,null,1963,1091],"b59274efd93ed87228f9d9ac846999797f7f60094b74a812dec15c9cfcd294e8",0]
+["mma_m16n8k32_f32_e3m2_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[1964,null,null,1965,1094],"0802de242b73d83cc7ad41a3b73a50833fbe7d5e03010e57e7a309b7600e6442",0]
+["mma_m16n8k32_f32_e3m2_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[1966,null,null,1967,1097],"11bb59d960f92c563f8ad65bc164854f8ac445a72bee421db20bab151c500d59",0]
+["mma_m16n8k32_f32_e3m2_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[1968,null,null,1969,1100],"f93ca756d143a8b3683ce162924d1e0c3e66c115b229c7f414778ca0ab25a522",0]
+["mma_m16n8k32_f32_e3m2_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[1970,null,null,1971,1091],"eb34d3235b7a837909c3dceeb5dfe78906bf6eb1230c9acbf99383dc6cc78fd4",0]
+["mma_m16n8k32_f32_e3m2_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[1972,null,null,1973,1094],"cec1854189674f20233806972791ea49c9c47fb2d8db6ec7c6e29fde47a1dd37",0]
+["mma_m16n8k32_f32_e3m2_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[1974,null,null,1975,1097],"3d73270231eb2c3ef546e7a99fb272c78752bab02a6cb1a3ce9df4149245638b",0]
+["mma_m16n8k32_f32_e3m2_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[1976,null,null,1977,1100],"5417ffd3b468e9bcccdb13a6786fee5938a63387e233a2bcc9df8b2d5def15a9",0]
+["mma_m16n8k32_f32_e4m3_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[1978,null,null,1979,1091],"3b531bc44b719b80217441b0a4199dc4eb5eb27554a7fcb7879aa21a31971069",0]
+["mma_m16n8k32_f32_e4m3_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[1980,null,null,1981,1094],"47874cde1835c158a995e9a4f73c86f728c8b7807c09e582cbcb2527d23b28ac",0]
+["mma_m16n8k32_f32_e4m3_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[1982,null,null,1983,1097],"3e87a0d0bc1fac7c8b4631e1658d627dbb5793cc7ae82858c1bbb33571fd8f18",0]
+["mma_m16n8k32_f32_e4m3_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[1984,null,null,1985,1100],"db91c7caa478ca7e7c4e160d6d56429235097165fc64ed59d1749668ba6713d4",0]
+["mma_m16n8k32_f32_e4m3_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[1986,null,null,1987,1091],"59774dfedb617cb077f045c7f91e5e96b4de4029ff9bcb262c4e2199d0833c15",0]
+["mma_m16n8k32_f32_e4m3_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[1988,null,null,1989,1094],"bb74eace92f9a7e8404e4b0c0ff03f92f8b1c58ffc1a3a77ca43fd0be28aed26",0]
+["mma_m16n8k32_f32_e4m3_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[1990,null,null,1991,1097],"92d7ac49e28717c877f1e00b57918332b38c2af871d867baa9fce1ec3a491477",0]
+["mma_m16n8k32_f32_e4m3_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[1992,null,null,1993,1100],"8267894d707e5a113ae7b90caca57ae51647663ee525f9808f1c2dc3ffe17e70",0]
+["mma_m16n8k32_f32_e4m3_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[1994,null,null,1995,1091],"98ce60c245e3f9ed147a42e5ce54b0a3e1e641d65b622c05bdfe9659b8bf0057",0]
+["mma_m16n8k32_f32_e4m3_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[1996,null,null,1997,1094],"42d5e07a6f28045d1823399f49dde55ebec252011ad56d2d95678777115e342b",0]
+["mma_m16n8k32_f32_e4m3_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[1998,null,null,1999,1097],"9f3a36e7f920dd2d79889642ed8f88ee1b19ea90097c19ea876236c71af9a832",0]
+["mma_m16n8k32_f32_e4m3_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2000,null,null,2001,1100],"c6377b0b1857b806a91ed5eb988a33afe129906377634c07ae8dbcce743b2b34",0]
+["mma_m16n8k32_f32_e4m3_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2002,null,null,2003,1091],"32bd99047eed48a6ce3d4677f17ff7fe0034edea530f96346bad26c04038780f",0]
+["mma_m16n8k32_f32_e4m3_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2004,null,null,2005,1094],"47ecb70273e5067eb410e23e79314f1d5f27f15e1456503adfe1dbe538b1a57a",0]
+["mma_m16n8k32_f32_e4m3_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2006,null,null,2007,1097],"87ee20c56f0a469f0d20eb9d7b706f667d18405e889a80db3ee7471e18575689",0]
+["mma_m16n8k32_f32_e4m3_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2008,null,null,2009,1100],"17943f81b7503c79e1b0632771672b4bfbe8fb596897d140370db3761f86bb46",0]
+["mma_m16n8k32_f32_e4m3_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2010,null,null,2011,1091],"e1ea9cbe86e29fcbe4fa99c440477918d2c986db69fedd3e59eb7b4a24ea2f73",0]
+["mma_m16n8k32_f32_e4m3_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2012,null,null,2013,1094],"59a2b377cb3c321d67f9bb09e40b054949368bac35379b3ef93996a788f8818c",0]
+["mma_m16n8k32_f32_e4m3_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2014,null,null,2015,1097],"0b1d4afe8a9d5f20368ad043f869f02056710a100f16cbda9cb10bab59881971",0]
+["mma_m16n8k32_f32_e4m3_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2016,null,null,2017,1100],"999669b04022ca36a4647b617f979af0dfd2a4973b23b6921a2228a2703da921",0]
+["mma_m16n8k32_f32_e5m2_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[2018,null,null,2019,1091],"8d96cae501fccc304856e10ceaf31d2606d9f802abd122a92379b4d7e0932eb9",0]
+["mma_m16n8k32_f32_e5m2_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[2020,null,null,2021,1094],"3186f84c03963837ec3e87a39daf2a698385ec75e3bf693f47423c4b1812d893",0]
+["mma_m16n8k32_f32_e5m2_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[2022,null,null,2023,1097],"6bdb38738edfc3644639853846f4b401fe1a46a1b50fed87fb4ee942b34195b0",0]
+["mma_m16n8k32_f32_e5m2_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[2024,null,null,2025,1100],"b103bb8bad65b0f501b1f6f53bb5f6e49556ed2f415600482ee7570c9e6ac7af",0]
+["mma_m16n8k32_f32_e5m2_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[2026,null,null,2027,1091],"487098b2fbaf042dcfe86d37b99289696151b3be5b9c8b5cd54def9502c0dd07",0]
+["mma_m16n8k32_f32_e5m2_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[2028,null,null,2029,1094],"e002a68faa229c95871fe991de9b49ee5db15ac98e43087d0e2143b0111fc8c7",0]
+["mma_m16n8k32_f32_e5m2_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[2030,null,null,2031,1097],"ef1ea4e064c62ec1e524b917a93a270f4965244468539a48944cd1dc79325f04",0]
+["mma_m16n8k32_f32_e5m2_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[2032,null,null,2033,1100],"9852aeea44d219e79512c4efe10dd06c4902b6e6049d7574590357b29d64dec4",0]
+["mma_m16n8k32_f32_e5m2_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[2034,null,null,2035,1091],"cb15124756b1245d8134630e6e0fc7c69e5dbcedd129f1da5e4f65e339058329",0]
+["mma_m16n8k32_f32_e5m2_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[2036,null,null,2037,1094],"5c6f345ce138c81a9f529d04b53677a24f84ab44f9eca85eca6445ea4988aa15",0]
+["mma_m16n8k32_f32_e5m2_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[2038,null,null,2039,1097],"4ff1f0516cad20e89814898490c2183f8239a5c0a4ea85ac8f94672170254454",0]
+["mma_m16n8k32_f32_e5m2_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2040,null,null,2041,1100],"853295a09403e49d094694345499ac4eb1387996590958072152337e483d98da",0]
+["mma_m16n8k32_f32_e5m2_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2042,null,null,2043,1091],"30dea76c1b44aec34aae2a46fdf4dd94aebe6359cc66d6dea095610af9514aa2",0]
+["mma_m16n8k32_f32_e5m2_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2044,null,null,2045,1094],"4e91298ce2cb70fab9b87db753e80dc074937b0d10e39b8eea204c3d2b416168",0]
+["mma_m16n8k32_f32_e5m2_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2046,null,null,2047,1097],"679ea73df260d9aaff60bff905e2688d4b56a3cf7c0bd98d8b48cf768ff9cdac",0]
+["mma_m16n8k32_f32_e5m2_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2048,null,null,2049,1100],"d4133e18e664fa4a7fe1d4230b5ddc0f25720024f931a12ba81fdbb04d93ae7f",0]
+["mma_m16n8k32_f32_e5m2_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2050,null,null,2051,1091],"3cc02cfd5817cb8a7be8c376d97995ba231d52248f29350230e690e7e277776f",0]
+["mma_m16n8k32_f32_e5m2_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2052,null,null,2053,1094],"a32448777a1663c46bb27d08c150b4e67a6b0144ea282a6859c847280a88e53a",0]
+["mma_m16n8k32_f32_e5m2_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2054,null,null,2055,1097],"3fd2ed9c746cdb27da8e9a36ac9816e438be9b3ec25b0708ba7d25496c58ab6a",0]
+["mma_m16n8k32_f32_e5m2_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2056,null,null,2057,1100],"64f7dd6cee62b810cc9791a00b9c5a7c730ea8262bcf337c387b1f4b5375872c",0]
+["mma_m16n8k32_fp8_f16_e4m3_e4m3",0,0,"sm_89","+ptx87","sm_89",[2058,null,null,2059,714],"d4d4c9d3b794b1a923b88b00831754c148d4189858d60d80860262002facb943",0]
+["mma_m16n8k32_fp8_f16_e4m3_e5m2",0,0,"sm_89","+ptx87","sm_89",[2060,null,null,2061,714],"13d5e739873c8e3c6208e0cc6972e76250e2730ca9baf457d3ba1e09241e199a",0]
+["mma_m16n8k32_fp8_f16_e5m2_e4m3",0,0,"sm_89","+ptx87","sm_89",[2062,null,null,2063,714],"74a7c00186b7ced8f6ce822c9363060a37f95e0a8059af75d7deb74c1bde7f8a",0]
+["mma_m16n8k32_fp8_f16_e5m2_e5m2",0,0,"sm_89","+ptx87","sm_89",[2064,null,null,2065,714],"59500a61a2861117814848e9dc7aa908f24d40aff7cc195303a5c0226f7b3c89",0]
+["mma_m16n8k32_fp8_f32_e4m3_e4m3",0,0,"sm_89","+ptx84","sm_89",[2066,null,null,2067,714],"e0a2361745fb072ed8b7ddc532234327410d258b36a303ec5e45eef94275fb82",0]
+["mma_m16n8k32_fp8_f32_e4m3_e5m2",0,0,"sm_89","+ptx84","sm_89",[2068,null,null,2069,714],"52cdcbbd10137c6e82fe4fc1901398e5768237407f169af70acb34b423b10e40",0]
+["mma_m16n8k32_fp8_f32_e5m2_e4m3",0,0,"sm_89","+ptx84","sm_89",[2070,null,null,2071,714],"09c469459c8c6a0b0e4e5eb2982cacd35403408576699568b8eb59606e23dc79",0]
+["mma_m16n8k32_fp8_f32_e5m2_e5m2",0,0,"sm_89","+ptx84","sm_89",[2072,null,null,2073,714],"3270172b1f4253bd2a85165e759cbd4ef93faf413fa0e86268c332f014dc8d96",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[2074,null,null,2075,1091],"e5053be64e9d04cdc70ad9b3d1afa77d287fa7c935c5a0dd1be50f7821868a07",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[2076,null,null,2077,1094],"f35ed6220c0f01b2f09f4d91fe7f11be26809e59fb623610052d6b6b04e37314",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[2078,null,null,2079,1097],"7f45779d78e34586da381a51866ffc098e2858f5551362d93e31c1617ad5e576",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[2080,null,null,2081,1100],"fca0d7c571339f2d8b304be8f2848004481610943a066cc3fd7aede546794a82",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[2082,null,null,2083,1091],"2ac662910bbead9c5b78de42633b7e7ad9202c36d791c7ab2474c8bdac3af30c",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[2084,null,null,2085,1094],"5dfb4df69f7b580084473674c25c3f50793eec088137ea4f5eedfa2306ecb71d",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[2086,null,null,2087,1097],"0fc6bb172bf870aab9ab845f76504060b44372ca7294196d5a6449702f91560a",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[2088,null,null,2089,1100],"c3fa148a0cff1c37818871785c86f7cb5118a9eff6bc4376bdcebe5b27edb8c0",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[2090,null,null,2091,1091],"96d81aa66f8691d4146c3dbbf038fb7daa8437e321c313c79244734896aba51a",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[2092,null,null,2093,1094],"e51254f0452d35c7b80d2762f6ecb36d4b5d528ea88bb7a80362a98d2032960e",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[2094,null,null,2095,1097],"f4e86139146a0f0fd6f8f7aba4078382ac7d088016a35c7ff3cd6309a127ee52",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2096,null,null,2097,1100],"aa396423f65e8a3ae41bea20defd9f642fc2d7f61791b664acffa1afddcc41ff",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2098,null,null,2099,1091],"04d23305255bafd393459e8cf75cda6a91e1a83f36d404e1cb33e2277f050d42",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2100,null,null,2101,1094],"4e7df1c787101ce41f8fe360de8cd4eea8155b1fe9418135b4e98f209cf3c599",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2102,null,null,2103,1097],"127448b6aa4f8a8b48b7b18039a04dbb49427fa21c4da1e6682e4c3771585ce3",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2104,null,null,2105,1100],"318f32c7cb4df62fbbefa3192243b8bd3f66c5816e25f806771691a45657e585",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2106,null,null,2107,1091],"d221ca7ded1586efca173f2308c9d90f7c3d197d1926be3055e682d7ce5d7e4c",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2108,null,null,2109,1094],"5127a9a8df3b355f9b83fc74a2c08dde3cb83d7bfc1366b69d872fc453116238",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2110,null,null,2111,1097],"23e02b6377f1132525b2f7a412b574ab13156c90d9304a10a345435a09356e80",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2112,null,null,2113,1100],"b52be672a4491cb4e1eb5faeeb6e5e1b6473a7508ec132f3849fd3fbedf41875",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[2114,null,null,2115,1091],"24a5f31cc538679fa85ee64c16d321dfba5da7785753421cb8e3903030751020",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[2116,null,null,2117,1094],"bff109fec5395947888a9da932ae4cfeaa9a1bf0e114c49df792c42a6fe32354",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[2118,null,null,2119,1097],"c508c60de6556cfef06a01ffa1bd7a59503f0f269584f603d6f5b72e944ef5c8",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[2120,null,null,2121,1100],"02a846335b86327046b985107374bb76a130ba607091f14cd05990a4e6ba9374",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[2122,null,null,2123,1091],"f137acaa2e864c44f92a59926333db34eabe5bfdec6c5b2c323abbf1a0e7e69b",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[2124,null,null,2125,1094],"6f064046a8cdf36ffeb3e3535159568d9142912ea89ede55c3e0817fd9e75773",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[2126,null,null,2127,1097],"468dc8dbb23e02f767c93aafbab64aba6473d267308d35a77d3527fd552f212c",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[2128,null,null,2129,1100],"4318146af334e3bb1d95dfbad102c0c64e061d568b80deeb196004891654c968",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[2130,null,null,2131,1091],"83e7c9c1e03f8fb06882565f081ef11b9e134037580486281bd5d4237ad2cb60",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[2132,null,null,2133,1094],"0ad12484cec2567b9d4e35fe124d02c045a5fe21459f49a29cd8620b84774190",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[2134,null,null,2135,1097],"62bd9e51ca2cd42b46968e323ef3206066835c4d22db06ca58c8eda79c03f605",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2136,null,null,2137,1100],"69d27e7fa610fb6756e2f73e983d7cfe77ae7f5d685cd08da801cb20f7701c44",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2138,null,null,2139,1091],"1bb08d8097575a52072c49f3e4567dd796d3e18ec0da5f586d4d4fb6f37767b9",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2140,null,null,2141,1094],"2edaf6149b4385d0d7457304fa7df64a9761cdd0c40e004455ea3ec343fdd8ed",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2142,null,null,2143,1097],"6116d8eca69739860b9c90cbc62a1ddd44ed9a76055343f848f2657c07a1f0e0",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2144,null,null,2145,1100],"d10b05cd7ce1b13c92203c28a74516b7ee318d06cc01cec7ba2b4c29d7dae397",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2146,null,null,2147,1091],"e2bfa84e117572019233f56d84425a311a85f80bebb6b11d6f52207355a2d6ff",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2148,null,null,2149,1094],"a6dac2894c4a4f0f2dc95e37eee0c7feddcbee22ac268c55083dc5e718f7f511",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2150,null,null,2151,1097],"3e2f8abce44f53d3c9a1619b832b0368409c048c2dc2658f144bed004b0aa4f9",0]
+["mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2152,null,null,2153,1100],"1c0ee2bcc55fa13d07001b18296a0889cc5874c45957a89bf00e7c8741ed69d8",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[2154,null,null,2155,1091],"b3173a2cf5305ffc9103294868836630b0ec6cc76f404aaccee483ca62e0518e",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[2156,null,null,2157,1094],"dc26455615aea42bcb65a15e3648797ef177b7c82dd5750431aa522a31e4c621",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[2158,null,null,2159,1097],"d535b6a2ed2250690855099fba7cf36a22e4c389845a828a6a31a44d4589f86e",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[2160,null,null,2161,1100],"3f958fde4f750db7374ae03dc857c5b3e305723b7da5c1cfc8cdd3a2ad0cb694",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[2162,null,null,2163,1091],"4fd5bb22c0bf5311959bd6e751178db035b094721c011d6990f99eeb03b76bfd",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[2164,null,null,2165,1094],"62e9bc37145f2a6ccddffdb23afba135c34b38f80447d31298cf20c899ee2081",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[2166,null,null,2167,1097],"1d215584f535280d6496bb73bbad714ab339c70f834ee103c302c2e21df6e948",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[2168,null,null,2169,1100],"191c292cbcd2e93b4b7e74b5195aa1ed63d1624fe5440ba5683d11ae9d0afd16",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[2170,null,null,2171,1091],"fb12999068ec79c74cf965e83653964e0d8432baadf5f8338c4ecbe0d580e15b",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[2172,null,null,2173,1094],"b762b268830a087e599f0737c06eb00c6738819a116f33dc425068e052a1aa6d",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[2174,null,null,2175,1097],"f9104de6075f24991ab0703b5cbfa17be74fcc6b4ed261eebf16fc0fee6df78f",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2176,null,null,2177,1100],"005299bcfe9d22162924743bbb8e3d1eab86a3fe6724a5abe9dbb3ee8acb4122",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2178,null,null,2179,1091],"806e33c92f72f6b0dd635a57aa7e6cb3b9cf5ecf137a326e3daf97a0f09dd020",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2180,null,null,2181,1094],"1fb3c67e1c291a16024bf26b205c404b024d2c8ccb007a539e95cc8d8d53de4b",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2182,null,null,2183,1097],"e13527d4853005cb2f14414069df69a3017b33b6bb1672735d1256d5e232a73f",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2184,null,null,2185,1100],"951d5e7e07d02b30fc3f2489ad6c25356835e70cd88727a438da59d8df52cd96",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2186,null,null,2187,1091],"ea5e12b2cd53e15a3fced485a362d39bb675a7fc2cd931d2ac5143ff1e4cca61",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2188,null,null,2189,1094],"b7e6e1695cb1ca3cd5dfb2ad56da64ac1425ad44fc5ed914bd98d5f0713d6310",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2190,null,null,2191,1097],"58babe61034f99b09140839e42899b9ac83b69e64d17497e16ca1120bcda398d",0]
+["mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2192,null,null,2193,1100],"b7845945f87c82fdc473f2cafebf79a4cfa2c694774a5d8f5316421564beaea1",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[2194,null,null,2195,1091],"6a2b11f924c407a80f80286652ded2c60c7d32d4c4451b45e3331ce5527a2cc5",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[2196,null,null,2197,1094],"668ab9115cd2ce16df0e7d997b0d146cec7bd8bf6cbbb9bf8fac5d3ebefb7ec1",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[2198,null,null,2199,1097],"2dc867f8ae66863cd29515633fc4e9e8cf3e77615a09764e1d133411a5fa81a4",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[2200,null,null,2201,1100],"0208ad1687a5c61ccef2bd98cd7ef183bceffb6d60dca90f57fbe3e4d85d0e3d",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[2202,null,null,2203,1091],"05a0a5736cef439e5e86697a3816c6cbba7f1f0f1185066341bb9c454a8d314b",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[2204,null,null,2205,1094],"76f14ead4a2237c917da1091f95f92f40b408f0486f5227f4f2443461d73b6fd",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[2206,null,null,2207,1097],"22188b8fd1968a629e331c79e6dfb35e3652706fea629b5403c371be7c81d629",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[2208,null,null,2209,1100],"b3ff5829afd4217953b750addbb212e017a5872d378c03bfcc05882b1032c559",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[2210,null,null,2211,1091],"aefaaed181f5e09a8fae1972d47f05646a7608bb1b3a329210521faaf3fdb1a9",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[2212,null,null,2213,1094],"115981835a2334294ea2e02ef917ebc1b3c2b79d795eea6ce174a3b108af78fd",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[2214,null,null,2215,1097],"b29a3ca034ca4a70e9633068c46b7c45fe24ad977d048f1270281f4e957ad2e9",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2216,null,null,2217,1100],"d2ad91a8eda8e1b1faa05133c76906831862b677ddad4970046b3ca22953d86b",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2218,null,null,2219,1091],"bd12a47d3c7a13f8561bd87ef15a2813d101562d0e19823f4862d360c1daefcc",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2220,null,null,2221,1094],"e5267311fa699872f6c4a33dd50b0d4441761550f4f5c91b764f268a1bb85b13",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2222,null,null,2223,1097],"076b4f81a490e0e586bef68d2a277a771150247d61e5584e6d2bcbbd51b415ce",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2224,null,null,2225,1100],"2eec66f6c9a39a5ae764822b31a32d7156a3ebebc0feabaeb8a140d7c1236d0b",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2226,null,null,2227,1091],"e1882d0a4d20da539a7256808cb8d42945df59023fd1f68d62ca07983c1636c1",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2228,null,null,2229,1094],"ed41f96883b9f8b810ac6591a4038e135ca4be384e039d7db62be93e6742f825",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2230,null,null,2231,1097],"8b5c7b8115d379888b830aa5d4b0361c9d57ee2e46fb341604ab9cabe236c2ae",0]
+["mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2232,null,null,2233,1100],"b2f4fc9a54c29901f36972bea31b4c75f35792ff827677c0d62886fdc46ef2fe",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1",0,0,"sm_120a","+ptx87","sm_120a",[2234,null,null,2235,1091],"53beb04afdae177a5de28ffc3a469f376ed1bbac12a3f07ea098d30590851d9a",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1",0,0,"sm_120f","+ptx88","sm_120f",[2236,null,null,2237,1094],"517e0abb01431a5255ac747f64acf1da21db792915c76c32e1af3aac63be0cba",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1",0,0,"sm_121a","+ptx88","sm_121a",[2238,null,null,2239,1097],"d404561123e475c4a7afb66abb69cf43ec809c68ecb20f768a6a963b9b66e855",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1",0,0,"sm_121f","+ptx88","sm_121f",[2240,null,null,2241,1100],"d41d05cc65e43d8a4d1bfc00d85349280ba0ff0216ebba9347f554c437ea1b72",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3",0,0,"sm_120a","+ptx87","sm_120a",[2242,null,null,2243,1091],"c7203bcfbda8e6ae5a8e5eb274c24ddd17b9efb416fb9161bab3b76d9669641c",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3",0,0,"sm_120f","+ptx88","sm_120f",[2244,null,null,2245,1094],"6617d9b8efe915c46112b309632e64907751de6cfd0933b5e1a8276668e65c97",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3",0,0,"sm_121a","+ptx88","sm_121a",[2246,null,null,2247,1097],"1a370a02567c1bb259f78184e3c64aec2fabd39f79a998bd3c0175dd358f7ece",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3",0,0,"sm_121f","+ptx88","sm_121f",[2248,null,null,2249,1100],"c2415a885015474708b2671106bcfd63afca42659f1ad4ef2e04126c2224b882",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2",0,0,"sm_120a","+ptx87","sm_120a",[2250,null,null,2251,1091],"f0cdffd263eb7a93e7f60effd0bdd35dca800a4f2de2435c72f98c70579b1d88",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2",0,0,"sm_120f","+ptx88","sm_120f",[2252,null,null,2253,1094],"3c2260d85cf3bd7a88876f0399ebe6b0a9f7d0f903f3ee2e81d2e54878404a11",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2",0,0,"sm_121a","+ptx88","sm_121a",[2254,null,null,2255,1097],"f606ae66285aa108e84888a1dff300528a35f3a98cf18264a9576fb69129e96c",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2",0,0,"sm_121f","+ptx88","sm_121f",[2256,null,null,2257,1100],"375cb765f88b13a9c4ab20ea8d9505c81d5cc236808513b676afb3f331965030",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3",0,0,"sm_120a","+ptx87","sm_120a",[2258,null,null,2259,1091],"efc3e1ed0bffbe2b504554ca5814fd4062b754e8ebd33190ee9fd2fde4cee423",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3",0,0,"sm_120f","+ptx88","sm_120f",[2260,null,null,2261,1094],"c83457bca70f158e36ae3a329cc46ace531885f8e3e9089a0d509dd7d8d74501",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3",0,0,"sm_121a","+ptx88","sm_121a",[2262,null,null,2263,1097],"a66652fe02684d87407e74971d726f0e594935ebb45a62b605bb0540a933d20e",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3",0,0,"sm_121f","+ptx88","sm_121f",[2264,null,null,2265,1100],"4a926a70db28493e9ddeab0b40c309f8f4fcbcef70044556530f668a7d605573",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2",0,0,"sm_120a","+ptx87","sm_120a",[2266,null,null,2267,1091],"7019eedc79a0c0059c5033436f3a475d645280ac4dca143e225742a601a1cc92",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2",0,0,"sm_120f","+ptx88","sm_120f",[2268,null,null,2269,1094],"706a0fec15fb081f8eab5aa69ab3300ba26a098d4e8d3dc303065b2237bb0c20",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2",0,0,"sm_121a","+ptx88","sm_121a",[2270,null,null,2271,1097],"453e880c3f1cd1854566f2ff634e642c6fa1c60440945214f4aa52e3cb714075",0]
+["mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2",0,0,"sm_121f","+ptx88","sm_121f",[2272,null,null,2273,1100],"50bc7ca01640d2b88e7ad4835a11824b3dd28db7ffa97b2300127d1eeaf4b1d6",0]
+["mma_m16n8k32_s32_s4",0,0,"sm_80","+ptx70","sm_80",[2274,null,null,2275,2],"1503119a63b21bf53b3da8e19a9f56ef74839eba90b286cb913fe22dce371031",0]
+["mma_m16n8k32_s32_s4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2276,null,null,2277,2],"a8ac7ca834182fd89087e3551e3f79bf3ea154e8b73fa0ddb0bc5b774013633b",0]
+["mma_m16n8k32_s32_s4_u4",0,0,"sm_80","+ptx70","sm_80",[2278,null,null,2279,2],"c999caa37e9bcbb9b67be4bef936e495172321c9567b5505949062f1704c6b50",0]
+["mma_m16n8k32_s32_s4_u4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2280,null,null,2281,2],"d491571d4bc24ae6003788dbd4449a384ef4b782b50df0d548ba63adb3e47b88",0]
+["mma_m16n8k32_s32_s8",0,0,"sm_80","+ptx70","sm_80",[2282,null,null,2283,2],"5836c7f30050fc5d819d945684c8375ef0534dfed9861862e2eb85178c76e48d",0]
+["mma_m16n8k32_s32_s8_satfinite",0,0,"sm_80","+ptx70","sm_80",[2284,null,null,2285,2],"1d85fbdf7370f70bf6eb5f9faa98bbbae97756dd3a5a6f2f004cabc2d31afa19",0]
+["mma_m16n8k32_s32_s8_u8",0,0,"sm_80","+ptx70","sm_80",[2286,null,null,2287,2],"4bcb66e2eacfff744a593bd1de0bcca03f049011009bde98a994411d6181abc8",0]
+["mma_m16n8k32_s32_s8_u8_satfinite",0,0,"sm_80","+ptx70","sm_80",[2288,null,null,2289,2],"a09464ec88ac8db05e7774480f53d53c18086e89f137badfc11dc621a449d3e3",0]
+["mma_m16n8k32_s32_u4",0,0,"sm_80","+ptx70","sm_80",[2290,null,null,2291,2],"0422e5726af709a70bc213e8288fd65567c4472f02b32c4bac2640a0c589a333",0]
+["mma_m16n8k32_s32_u4_s4",0,0,"sm_80","+ptx70","sm_80",[2292,null,null,2293,2],"fb67b040d819991604762d4a3eeaf95c4a7f29a6a9c6fb7c033712953e0b967f",0]
+["mma_m16n8k32_s32_u4_s4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2294,null,null,2295,2],"e455468fa6a140ec116a333be98f103a12a1c5b5b1402c85ff9dca72d49d39e4",0]
+["mma_m16n8k32_s32_u4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2296,null,null,2297,2],"6723714b7973b080e2a377280405b00299fb0124d14eb50a6f26749bdf2be45a",0]
+["mma_m16n8k32_s32_u8",0,0,"sm_80","+ptx70","sm_80",[2298,null,null,2299,2],"c32aeb79cafcb949caf6cb4042656ecd6e2cdc557b32c08cf66998862dbe2646",0]
+["mma_m16n8k32_s32_u8_s8",0,0,"sm_80","+ptx70","sm_80",[2300,null,null,2301,2],"3fe0f1795262b7943b8f9f87d8a322986293a2c8b5bda81c803cbf1a833022ff",0]
+["mma_m16n8k32_s32_u8_s8_satfinite",0,0,"sm_80","+ptx70","sm_80",[2302,null,null,2303,2],"20ecf677a324107e442e86c16a8c38c8a8e1fc76b7d23d99cd92e1e837b74fcc",0]
+["mma_m16n8k32_s32_u8_satfinite",0,0,"sm_80","+ptx70","sm_80",[2304,null,null,2305,2],"53c3969c29713177c61219e45eb25f451ae5c85e5cdfd827618d6041ec9b42f4",0]
+["mma_m16n8k4_f32_tf32",0,0,"sm_80","+ptx70","sm_80",[2306,null,null,2307,2],"18d11394651fa6467d023c0e71c9177f708681aaf13d2ce5e27a0b1969b19ec1",0]
+["mma_m16n8k64_s32_s4",0,0,"sm_80","+ptx70","sm_80",[2308,null,null,2309,2],"0072518b09527167a165e0bc45c3ad5007bcd7837041bc87c234495b10fad0c9",0]
+["mma_m16n8k64_s32_s4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2310,null,null,2311,2],"6ff231d4a92e0dd725d769fe30a4762fb774f6e5030311e6f9daee55e7cabc6c",0]
+["mma_m16n8k64_s32_s4_u4",0,0,"sm_80","+ptx70","sm_80",[2312,null,null,2313,2],"967bc49e996fbeba225b2c48623201817f0a46d3d8517b781fe3732593cb8a54",0]
+["mma_m16n8k64_s32_s4_u4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2314,null,null,2315,2],"0776a1700bbed9f71aae1fb81c5b20258db17a40a3d64e0a1d8af9b67f06f0fb",0]
+["mma_m16n8k64_s32_u4",0,0,"sm_80","+ptx70","sm_80",[2316,null,null,2317,2],"a53378089d4cfd0264fdcee767b40b2e12805f1e9554d864081b531557f9a9af",0]
+["mma_m16n8k64_s32_u4_s4",0,0,"sm_80","+ptx70","sm_80",[2318,null,null,2319,2],"91d89d23fba8c8d418f6dc699d482d5d1019705df5633fc1bd747334607adf55",0]
+["mma_m16n8k64_s32_u4_s4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2320,null,null,2321,2],"404ccd4d57dedb0b5fb5f2ad39657c89cdf07498ca5c9bfb1ed53a7b77e8aedb",0]
+["mma_m16n8k64_s32_u4_satfinite",0,0,"sm_80","+ptx70","sm_80",[2322,null,null,2323,2],"2d5f832f846391c5c9382e87c9b273c2e9b8a41cfb61b582b80fa6b8eb0bd229",0]
+["mma_m16n8k8_f16_f16",0,0,"sm_80","+ptx70","sm_80",[2324,null,null,2325,2],"9ba7139ff869de74591d13393818db59ba86726fae9ee0dae9991acb6da3d5b1",0]
+["mma_m16n8k8_f32_bf16",0,0,"sm_80","+ptx70","sm_80",[2326,null,null,2327,2],"91f88f86c974979c340247a2024115e679dde1feaa79166b262f6bf03af76156",0]
+["mma_m16n8k8_f32_f16",0,0,"sm_80","+ptx70","sm_80",[2328,null,null,2329,2],"19eb7e71e03a6049eec2c7e7fa3e296b904bd9ac4fcbf641151468c63f27322a",0]
+["mma_m16n8k8_f32_tf32",0,0,"sm_80","+ptx70","sm_80",[2330,null,null,2331,2],"f096329555d7d8b9343218e7b3a3cf76c84a2997a2ad0accf71c8e615138b7ee",0]
+["mma_m8n8k128_s32_b1_and_popc",0,0,"sm_80","+ptx71","sm_80",[2332,null,null,2333,2],"fda7fb129fd283b729840242e4a672b235fb392188cef3a412b70b2faebd5be2",0]
+["mma_m8n8k128_s32_b1_xor_popc",0,0,"sm_75","+ptx70","sm_75",[2334,null,null,2335,7],"67ebfe84365b42e3abfa551ff04c084c79324c2d948728ccfd1fb6b26b22fa4a",0]
+["mma_m8n8k16_s32_s8",0,0,"sm_75","+ptx65","sm_75",[2336,null,null,2337,7],"6e445760743135e869e510b30111f5ccbaf5deabcf06c1c8f5898fd5433e5682",0]
+["mma_m8n8k16_s32_s8_satfinite",0,0,"sm_75","+ptx65","sm_75",[2338,null,null,2339,7],"09d07af817b846c004f8f82e2565bc2d85e6c9c8e3637f4edc4331e73c8618e5",0]
+["mma_m8n8k16_s32_s8_u8",0,0,"sm_75","+ptx65","sm_75",[2340,null,null,2341,7],"31766cd77392c23a4e0b4fcbe85ce19692b6e65f32a0cef25fc3a0b6951a609d",0]
+["mma_m8n8k16_s32_s8_u8_satfinite",0,0,"sm_75","+ptx65","sm_75",[2342,null,null,2343,7],"36f97c64cd5522330a2bf069f8d06bf82e2aaec64fb231a6ff89f50c59ef97bb",0]
+["mma_m8n8k16_s32_u8",0,0,"sm_75","+ptx65","sm_75",[2344,null,null,2345,7],"b0323ca11023c1216e989adf1964e081b00d6ebfe1dba53c4fc666572c84a2bf",0]
+["mma_m8n8k16_s32_u8_s8",0,0,"sm_75","+ptx65","sm_75",[2346,null,null,2347,7],"2c6f43c42ec6e86021f0956e38222ea70270c33f8e04d4e2a565e3f398041a1e",0]
+["mma_m8n8k16_s32_u8_s8_satfinite",0,0,"sm_75","+ptx65","sm_75",[2348,null,null,2349,7],"e7175f04bf707d76df911ef86029960a8e5405f5b4e42cff2e44097af0f8bacb",0]
+["mma_m8n8k16_s32_u8_satfinite",0,0,"sm_75","+ptx65","sm_75",[2350,null,null,2351,7],"cbec956f7d3c53684f7293fd1a8ba0ec02572f117bf3125e52e4a19b0ec8989b",0]
+["mma_m8n8k32_s32_s4",0,0,"sm_75","+ptx65","sm_75",[2352,null,null,2353,7],"688b7ba7fae3ca5e409d9a3410fcc411bfe8fd77948232f783e0a62d99271b6b",0]
+["mma_m8n8k32_s32_s4_satfinite",0,0,"sm_75","+ptx65","sm_75",[2354,null,null,2355,7],"54321dda2dfb6361116ce1ab6362e9dbbb293d13453f821b3ddd6bb89f143e1e",0]
+["mma_m8n8k32_s32_s4_u4",0,0,"sm_75","+ptx65","sm_75",[2356,null,null,2357,7],"9732613487d40eb02d26fa3e8200ed55af7f3bb37478a6b38f4012591cb9116c",0]
+["mma_m8n8k32_s32_s4_u4_satfinite",0,0,"sm_75","+ptx65","sm_75",[2358,null,null,2359,7],"4fa65472e5476f333b2b993ec2efdfdeebf6d3ab0a97f439b4cb2631f8edf220",0]
+["mma_m8n8k32_s32_u4",0,0,"sm_75","+ptx65","sm_75",[2360,null,null,2361,7],"f930f37e2a096810fa620c0ea804dd89ab9b192db1040ee8df798efa1d5a09e1",0]
+["mma_m8n8k32_s32_u4_s4",0,0,"sm_75","+ptx65","sm_75",[2362,null,null,2363,7],"11fedb7d711342cd9671d94d0fc7189ac6ed8c2b6705bf139a13b7d439cad8b9",0]
+["mma_m8n8k32_s32_u4_s4_satfinite",0,0,"sm_75","+ptx65","sm_75",[2364,null,null,2365,7],"3cbc09a59cdda988e33747fbada295b1d458c507b0e6fc1f6a21eb6bef846af8",0]
+["mma_m8n8k32_s32_u4_satfinite",0,0,"sm_75","+ptx65","sm_75",[2366,null,null,2367,7],"09a2f56e47e1b1433fd8680f23702e3c9a924feb93b00bc914277daf7148218c",0]
+["mma_m8n8k4_f64",0,0,"sm_80","+ptx70","sm_80",[2368,null,null,2369,2],"ebec9fc9bf431dcbe8ee681ee78bcac4a76236267c7386f715b874db09f662c3",0]
+["mma_sp_m16n8k128_s32_s4",0,0,"sm_80","+ptx71","sm_80",[2370,null,null,2371,2],"7cad7817965b0ced4932e71dad3541d58b4f850264a9d0e9dbe7ccf7f38eeb2e",0]
+["mma_sp_m16n8k128_s32_s4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2372,null,null,2373,2],"ed815c72c9d76f6e6d7f8eea9063496364e9de47c97741ee77e0b87c8d86c8ee",0]
+["mma_sp_m16n8k128_s32_s4_u4",0,0,"sm_80","+ptx71","sm_80",[2374,null,null,2375,2],"066560a9b844f214de0c498348ace22e2c1c654800b4f783620973c18bcf4e6d",0]
+["mma_sp_m16n8k128_s32_s4_u4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2376,null,null,2377,2],"2e13be2390d4bff14f1caf99581d9668b54d3904b96bab380c67316a433f65b5",0]
+["mma_sp_m16n8k128_s32_u4",0,0,"sm_80","+ptx71","sm_80",[2378,null,null,2379,2],"8b265775dbf53c60f2673d546730fab955a682d7040e8deb82f00388f7a9b066",0]
+["mma_sp_m16n8k128_s32_u4_s4",0,0,"sm_80","+ptx71","sm_80",[2380,null,null,2381,2],"e721dec5eeed0bcd286c96834b73b84a231257cf0d7c89490aef7204ef7464d6",0]
+["mma_sp_m16n8k128_s32_u4_s4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2382,null,null,2383,2],"748c450bd242e77e47ffe5e64ace2ddae83a3be8a8dacf9898c003c32d08cbc0",0]
+["mma_sp_m16n8k128_s32_u4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2384,null,null,2385,2],"ef0ae102a6833ec596b82870c5069906830afc86c36f8f3e281289280800ca08",0]
+["mma_sp_m16n8k32_s32_s8",0,0,"sm_80","+ptx71","sm_80",[2386,null,null,2387,2],"c13d5f8e5dcfe33a56677bb4285c2ae13add47fa6908bc5da850b4e1a8e918d5",0]
+["mma_sp_m16n8k32_s32_s8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2388,null,null,2389,2],"86a2c43c75df9a5e9508f41f7ee59711d51444a946b438368a01d240859aef49",0]
+["mma_sp_m16n8k32_s32_s8_u8",0,0,"sm_80","+ptx71","sm_80",[2390,null,null,2391,2],"d616527d469bae7f1e553054d369f155464f77663fdb6533f66a40635c8a41a6",0]
+["mma_sp_m16n8k32_s32_s8_u8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2392,null,null,2393,2],"406ee7477622e3d510c1fc43e78dd71b5b4f9ee5dabe911cbdf67dc4b7fe8683",0]
+["mma_sp_m16n8k32_s32_u8",0,0,"sm_80","+ptx71","sm_80",[2394,null,null,2395,2],"7b03c57dcaca632290fde77c68fd91148e74e7feb9503e2a1414b87ff470658a",0]
+["mma_sp_m16n8k32_s32_u8_s8",0,0,"sm_80","+ptx71","sm_80",[2396,null,null,2397,2],"7aef6d4fa52513fee635b89292fad6bcae7ac0be6f9a0ccc2256e22d27d76ee5",0]
+["mma_sp_m16n8k32_s32_u8_s8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2398,null,null,2399,2],"3f772532db677b46ee2d90350bb8f2e304166101ffab96b3733e435d04c006a9",0]
+["mma_sp_m16n8k32_s32_u8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2400,null,null,2401,2],"e5e27bd43925dab5a3c8363bb62b871561c89b814319ea44b510ee9004519c72",0]
+["mma_sp_m16n8k64_f32_e4m3_e4m3_f32",0,0,"sm_89","+ptx84","sm_89",[2402,null,null,2403,714],"eb4eaa13fce0d0944d19bd078e42c2e8407f4e193086d491355cae1e16e6920e",0]
+["mma_sp_m16n8k64_f32_e4m3_e5m2_f32",0,0,"sm_89","+ptx84","sm_89",[2404,null,null,2405,714],"83aed1212e0995fdd2b932524293ded5754e530b7111093c583f662829f46747",0]
+["mma_sp_m16n8k64_f32_e5m2_e4m3_f32",0,0,"sm_89","+ptx84","sm_89",[2406,null,null,2407,714],"e135b46a04ca22298acf3104c95911c77bc22660c5c2b40f505450299e1ea685",0]
+["mma_sp_m16n8k64_f32_e5m2_e5m2_f32",0,0,"sm_89","+ptx84","sm_89",[2408,null,null,2409,714],"3ce5c8074e6eb591de550ba1779b353496025212da6cc8a7d39f2a2dc706c55e",0]
+["mma_sp_m16n8k64_s32_s4",0,0,"sm_80","+ptx71","sm_80",[2410,null,null,2411,2],"5b7a0001c8de944e875321c6d5e7ade0210cfbc94c4cd21ff9cf65952d8f2cf5",0]
+["mma_sp_m16n8k64_s32_s4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2412,null,null,2413,2],"0bc2b70c067dcaa939ec1e8e56bc7c95716bdd2d817385a9861ce10d785e7462",0]
+["mma_sp_m16n8k64_s32_s4_u4",0,0,"sm_80","+ptx71","sm_80",[2414,null,null,2415,2],"b07becbd37bcf0af8a78f84f432128e266874b3a8dd59eb7415d15725559881d",0]
+["mma_sp_m16n8k64_s32_s4_u4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2416,null,null,2417,2],"c5fe58f54bdb02da6944b08860c134018a0de4cdd0047540319d91bc9a3049a6",0]
+["mma_sp_m16n8k64_s32_s8",0,0,"sm_80","+ptx71","sm_80",[2418,null,null,2419,2],"6bcadc0167c577370069f06809624977d564a96636e99f06f2f3cca32674bdfe",0]
+["mma_sp_m16n8k64_s32_s8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2420,null,null,2421,2],"054b5e6a36a6b619feb69ee699209395e60b2a3f513be892136a0fd45547ae2a",0]
+["mma_sp_m16n8k64_s32_s8_u8",0,0,"sm_80","+ptx71","sm_80",[2422,null,null,2423,2],"817db200dd96b3145363235c2db1dcacaf77f83af811f987c4e10e7817bb9f71",0]
+["mma_sp_m16n8k64_s32_s8_u8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2424,null,null,2425,2],"611f68e62fcd3832710faf0257b48d64f65cf3741d01ad8820293a255343a755",0]
+["mma_sp_m16n8k64_s32_u4",0,0,"sm_80","+ptx71","sm_80",[2426,null,null,2427,2],"acc681b9d01c80d41af7d23489d53dec56a6677830b33a18a140d7d81a0c946e",0]
+["mma_sp_m16n8k64_s32_u4_s4",0,0,"sm_80","+ptx71","sm_80",[2428,null,null,2429,2],"975a2ba30bbdf40639d4dbf47a1d1ff644f42e6f985a3a709dc698092a5e8df4",0]
+["mma_sp_m16n8k64_s32_u4_s4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2430,null,null,2431,2],"ab12bae7f950a57d55f53bdac98e1ba496b92bc62f244bc000dbfbe7e35e9138",0]
+["mma_sp_m16n8k64_s32_u4_satfinite",0,0,"sm_80","+ptx71","sm_80",[2432,null,null,2433,2],"230a40802371e88ec50e3d35673ab8e14c40bf5d36e3de9166f559b8cc9ddae9",0]
+["mma_sp_m16n8k64_s32_u8",0,0,"sm_80","+ptx71","sm_80",[2434,null,null,2435,2],"23f3e6fdcbc78049b2c4d63c0abbb57d398a180e9e7d016efe0444d61aa13f34",0]
+["mma_sp_m16n8k64_s32_u8_s8",0,0,"sm_80","+ptx71","sm_80",[2436,null,null,2437,2],"5f27ec1135fe5612e82a9fe10c733263f0b017aee9f9f439094460d1a9841669",0]
+["mma_sp_m16n8k64_s32_u8_s8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2438,null,null,2439,2],"2137e1cd758ecf78b828ee8aeaf486e3bb5702fd824984192fd89768653ef782",0]
+["mma_sp_m16n8k64_s32_u8_satfinite",0,0,"sm_80","+ptx71","sm_80",[2440,null,null,2441,2],"46d62c6f2450689a260c486061a0cdf71990ee76d4b5ba4607a6df9a8a4b89ff",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_s4",0,0,"sm_80","+ptx85","sm_80",[2442,null,null,2443,2],"8e282bf56d6b856321999b4eed804579a82b7821cc8de1c6137e0f81a2f50c21",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2444,null,null,2445,2],"62d98e2a44b7ad3e93b948cc131edd8125c4fcdee08bd669a74e3206b12d37f8",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_s4_u4",0,0,"sm_80","+ptx85","sm_80",[2446,null,null,2447,2],"37fe6a5e43c93c2f479435ff6a5e8b4b3f1d94a5d63b2f8a0cf2bb46226f68b5",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2448,null,null,2449,2],"ad96a6363ea670e47c6b900226b51b8a12e3f5a45f9c392fb4ef7430fc26aff5",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_u4",0,0,"sm_80","+ptx85","sm_80",[2450,null,null,2451,2],"3ea4fe7351963e05074a4b6d19fbe371642f4878eb534de6081fa98cfcd795ea",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_u4_s4",0,0,"sm_80","+ptx85","sm_80",[2452,null,null,2453,2],"decebf782da9d8709870f892b1a2be3052f868e7bea114fe295b5e0b72704cb1",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2454,null,null,2455,2],"f17ea2158d7dc4c408015d845b99a81e14d9b624050bb948318f10e9fd98fa80",0]
+["mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2456,null,null,2457,2],"fd7dd72a3b00d4345f194aabcedbf8629b8fe530ef1f5d028ce3b6651de65836",0]
+["mma_sp_ordered_metadata_m16n8k16_f16_f16",0,0,"sm_80","+ptx85","sm_80",[2458,null,null,2459,2],"fb770f39fc8ed1657857832ba951473f7e48d4e2491494e11bdb0026ddf5db7b",0]
+["mma_sp_ordered_metadata_m16n8k16_f32_bf16",0,0,"sm_80","+ptx85","sm_80",[2460,null,null,2461,2],"0caab71fa2c2ffa17eb8e3f6d5c678f8076bab1db812f16250c67f38694b1dbe",0]
+["mma_sp_ordered_metadata_m16n8k16_f32_f16",0,0,"sm_80","+ptx85","sm_80",[2462,null,null,2463,2],"33af663545cc25f6d113c039d7861a5715754fc775db4b2480265e3f36b2ab0e",0]
+["mma_sp_ordered_metadata_m16n8k16_f32_tf32",0,0,"sm_80","+ptx85","sm_80",[2464,null,null,2465,2],"acbe57fcb955dead69bc24ebc4f46579d1f36570acf0abd927cc854d77ca4d2c",0]
+["mma_sp_ordered_metadata_m16n8k32_f16_f16",0,0,"sm_80","+ptx85","sm_80",[2466,null,null,2467,2],"994665d1e1160870c47a2c629bf85d47177038c7f5366e94358bb89700effcc7",0]
+["mma_sp_ordered_metadata_m16n8k32_f32_bf16",0,0,"sm_80","+ptx85","sm_80",[2468,null,null,2469,2],"92797f1d3dabcd39841efc40d075c55227eaee08438c078dd2000d02f0539b5a",0]
+["mma_sp_ordered_metadata_m16n8k32_f32_f16",0,0,"sm_80","+ptx85","sm_80",[2470,null,null,2471,2],"cc5f52246275151f0dbd153dcbb59aab4c272aa0f095e35996fb0731c644540d",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_s8",0,0,"sm_80","+ptx85","sm_80",[2472,null,null,2473,2],"26621133d8c27b71b181047e2dd76f3363ad2f61f36bde9c46821f18a0df7a57",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2474,null,null,2475,2],"8699ce1c69b48d90d09fdbee8924f5d6881c2021b8a23645e04ed454f3d29e2e",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_s8_u8",0,0,"sm_80","+ptx85","sm_80",[2476,null,null,2477,2],"d5fafd43914fe57cb7f046f7871a2368724693b23ce926e1cd2d8296ad1ab845",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2478,null,null,2479,2],"84684a4102a54f29649724bd8f6c42fcfa015ddf25f4deb955a70f64998296b8",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_u8",0,0,"sm_80","+ptx85","sm_80",[2480,null,null,2481,2],"b65e2d19252cbed19158834df715faab5fd03535d11e9cef406d2340d157f432",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_u8_s8",0,0,"sm_80","+ptx85","sm_80",[2482,null,null,2483,2],"e7593ef82465716a8c5f99b508119ef9c2d2de1673eaf526d5f552bf2598f985",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2484,null,null,2485,2],"07ac5cd8c3fed9824c1be745c7c071583eeeb49613c4987317f541f9cb8e6b0d",0]
+["mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2486,null,null,2487,2],"398161353fbecd7bd63af781e7c9b4f510b018ec8f32c4b3aa10ab969066d2f6",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16",0,0,"sm_120a","+ptx87","sm_120a",[2488,null,null,2489,1091],"0c1fa31095e6cbfe9714fc22bc2c6baa4bed27b50ea7622742b4552b292650c0",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16",0,0,"sm_120f","+ptx88","sm_120f",[2490,null,null,2491,1094],"e323c290ee5713001279fdf1ee2fdd7798f835a3ed93216984c8a753b6fcd2e2",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16",0,0,"sm_121a","+ptx88","sm_121a",[2492,null,null,2493,1097],"68f668481d7b179be02e3403cc4ac3f6b5c02aa167e4d6b0969ac2609c63a6ee",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16",0,0,"sm_121f","+ptx88","sm_121f",[2494,null,null,2495,1100],"412718226922e7b43ef3e7fccebb6ce03208002dfa01dc3d9d6d08a2e71c5498",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2496,null,null,2497,1091],"67c8e057feeb585325d67cba1c4eddbb0017d93b91c90be0a9214140079f64d2",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2498,null,null,2499,1094],"2ad46501b8a1c461690b08f932f8bb4a76b4fb13ef337744855b5e3c7b8ba1da",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2500,null,null,2501,1097],"b0af1bb0262f7f439810f5190253139918dcf0b20fdbcf9600e2d2badfc5db8f",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2502,null,null,2503,1100],"f532438eec74197d55167f571f4fad612a7b1516f069ed7721057cd580f01284",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2504,null,null,2505,1091],"45ab878ddfdeb57e8cd3d16381f6bc7041405ee8af204530bc152939ffaa8b96",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2506,null,null,2507,1094],"af9508bf858d7988297ce2bff72309d2da862dd08bef377782f57208cb1d4491",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2508,null,null,2509,1097],"2309adceea13743b9aa948b84ff7d0e3438fb997eda647b8297a06cf8eb3f238",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2510,null,null,2511,1100],"4ad6d593bc21a603847d5c36efdb39aa83003609515d3e47ddb7d9a693a50092",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2512,null,null,2513,1091],"cc0638ddf5304df501211e10e6040171ee4be78e30a97c5ca943dd3b108f982a",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2514,null,null,2515,1094],"6096f4e808dd8e25cb81deda78c6ec4c1050440a985b9d5e63742f843dbfebb2",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2516,null,null,2517,1097],"38209bc8ed518743bf97c724581a43b8d878e45d43a8579aaa2c2f00f5572b69",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2518,null,null,2519,1100],"e297f531e48a946b8fa0ac6987e9227510600b8bf6387ecab5e0b7bb2e60b405",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2520,null,null,2521,1091],"bf0b2121eb5db0593712224b6cd0496f2863d9de5ca41ca655f608b54f5dc165",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2522,null,null,2523,1094],"af4f34a47f77ecda586237e17b025cbe2bee94896b1802c1be9b3e9eeb8ce743",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2524,null,null,2525,1097],"99f83cee2daa598c3e0a35969d19cf41ca307fd08253c60a3a20c01f7a4bef05",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2526,null,null,2527,1100],"a1fe16f839672ea57ae22efcc6cd812e7a75e0cb108c34239166a25d14339d3f",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16",0,0,"sm_120a","+ptx87","sm_120a",[2528,null,null,2529,1091],"d1627a7cab662e1ba2bd19c31e9f80ee469d1b8a2927f52b6489004afb557694",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16",0,0,"sm_120f","+ptx88","sm_120f",[2530,null,null,2531,1094],"6d49921c243b0c36654d60dd8f3e182f3c55204e548158ef0ae15f3bedca78bb",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16",0,0,"sm_121a","+ptx88","sm_121a",[2532,null,null,2533,1097],"3e33428b9f890bbca2a9dbb5541502cccb4a7cea31f7a579cdffc9bb442b0e98",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16",0,0,"sm_121f","+ptx88","sm_121f",[2534,null,null,2535,1100],"a04947bea5b136a7b2e62343430cfde276a617a4f037fc6b4a621165a49af7c6",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2536,null,null,2537,1091],"c499bb460958560032dd186801067048eb2c3765a41ca4c4f5a4db37853ac2d0",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2538,null,null,2539,1094],"80bd96b2c3f0dfd4a31c077e6db0a0b43de218079e250a1d682a0949026eff1c",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2540,null,null,2541,1097],"fa73d73e345f3df100e40574d6a891921b992efa53e9be12c5f107bedb866b20",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2542,null,null,2543,1100],"b43862394cbb6b2d75b2d025809d1326a614079f889c3bdd3a96c6ffa7771436",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2544,null,null,2545,1091],"446b60b57abe7d042c03f4cc24cdb7c0bfbf79e598b8baed6c2e52a908e05f3d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2546,null,null,2547,1094],"bfa01cb08cd38918c80dda5f6c97048cb9ce2fed79b4155966fa8252fd9a072a",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2548,null,null,2549,1097],"cf1762c88b137aeb0610dbe3b4ffc96cfca1f80ea0d3d4d6ce9caa09e01b61a4",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2550,null,null,2551,1100],"e967c10d018019f6db0425d97a1d83b10fb0c7cbee8b4c0866c01d07a955fca0",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2552,null,null,2553,1091],"49470eb5271826b54e96b55cd5e097fd1991f70d76d5bdbe7d9837d2f1198f95",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2554,null,null,2555,1094],"093c341f4d53f31ff155a28599fab645139d09ba12fc3ed7d97795a23bb0e082",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2556,null,null,2557,1097],"cb785a81d3e121217e9f7dc09d24d8aff35627eeb02411d434a38f762c189d2b",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2558,null,null,2559,1100],"6d861cd0666515700e55dc14cf70209789fa2b3193d87b09c80a05bc01673cf1",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2560,null,null,2561,1091],"4476babf2e20f43935bcc202ed0fc3deb6430de2d062c517dbfbecacceb695c8",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2562,null,null,2563,1094],"8e882838f8c88689d9b5d053178946a482c48b1bec530a7ae552e00958b52bd3",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2564,null,null,2565,1097],"b7cd1dc61cee306457549cdaf762a6b915eebd19ee8baf3f06eeedeab1c2a9f3",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2566,null,null,2567,1100],"8dc72e5dceab8f14651bdbbab6ed3453a6b16079f51be20be47bbc8de47ff0f8",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16",0,0,"sm_120a","+ptx87","sm_120a",[2568,null,null,2569,1091],"6d77ad29c1b813001c6e798842ab7c79b43f04cc8acfd051b019b69417873f99",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16",0,0,"sm_120f","+ptx88","sm_120f",[2570,null,null,2571,1094],"a4cb7275ce993631cba0d96ffa213b79d66109d0d48b1abc5bc44bf0d68f8b2c",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16",0,0,"sm_121a","+ptx88","sm_121a",[2572,null,null,2573,1097],"abb3a08e335043e9a64bfd074eca58116a24a22cfa0256d7ea0c24ae20a90e32",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16",0,0,"sm_121f","+ptx88","sm_121f",[2574,null,null,2575,1100],"26b7f97697bd07f865fb52d25b7f9b0a1158c9e14aecb99a541837cb7f657bae",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2576,null,null,2577,1091],"89eba35e54dd72f179ff1ab885b490594c5f1a70563334f37fdd4b9d4ed7c3aa",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2578,null,null,2579,1094],"4fd02945caba6dcda757571309c6a7f7ec0249c5fdcd6e55ad744b5a9522120d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2580,null,null,2581,1097],"3ffe526b6be471db6e4c92884118657223dfe5568dffb58799b4a29dc0a9f3cf",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2582,null,null,2583,1100],"9cadfb060c4ea97d11b9cec047a556c10ef551babf7f5ea78401ae9af79746ea",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2584,null,null,2585,1091],"8ec1403fdd0daed18fed6f188c459138e0bf388ebba6363ec7f062fd9742fb05",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2586,null,null,2587,1094],"4775dcb5a4bc8d2505fdcf4459f5a8dc6cc28acd5efc21d948f7838705a5c317",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2588,null,null,2589,1097],"9542b8e86f94a0a0dcb96d66e304f55fc26091bc43d3b37976441835398ae476",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2590,null,null,2591,1100],"72725b3ec7996595911bdb3af58ae511bc01c841fdaaddeecb14712c98f787f1",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2592,null,null,2593,1091],"6f0dad5633a7b062039848e95e2a93fb02b138540e15caaf10e3f2b69f7280b7",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2594,null,null,2595,1094],"e35bb773784698708eff304b8a847ac70e844da8bbf50d1d4676d82fa53a1c3c",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2596,null,null,2597,1097],"0080437a66c28896100abf4a816d7aa13413949603b21c376cdfd83b646ddd7f",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2598,null,null,2599,1100],"9141fff6df8aef6ecab0afdba9f0bdef9604225dc6224c158fc9c3cb9b309301",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2600,null,null,2601,1091],"76a6e0e241fc359debd042bfe0a1ed0d4f90142ea589dd5e6c7cf6728ab20be0",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2602,null,null,2603,1094],"464bc2d89f6590b9b36b6baaf90d3e217302786391f59898c5b2eea4f120d058",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2604,null,null,2605,1097],"e06a512848d89b67b4fb609c40045cb4c4d0f0d854d13070a1557064303b3a70",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2606,null,null,2607,1100],"db526599aac0e6379d8dc4357a95d31a9c88b9874f638fc185c963c4a5eaa016",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16",0,0,"sm_120a","+ptx87","sm_120a",[2608,null,null,2609,1091],"2afeecbcdae8644a7950c28549eb7b9bba0e9794f03b7a1efe3b92eb91e24fb6",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16",0,0,"sm_120f","+ptx88","sm_120f",[2610,null,null,2611,1094],"e6fc4d997aae73a8619ccf4a5b714acb318f04f8042611e1961b6a294571f830",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16",0,0,"sm_121a","+ptx88","sm_121a",[2612,null,null,2613,1097],"adc5493b4d95096dc1d224ea0147feb18f598e47f058e1c9b3f76b4feef72104",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16",0,0,"sm_121f","+ptx88","sm_121f",[2614,null,null,2615,1100],"0dea23eba0f490b07a6c66dceeea235f6d25e1f4be4f6782fa0c5d7a1d6a31b1",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2616,null,null,2617,1091],"de53037a0aa32af39659c25749ad2f6ae4d2a412e74ed27ec16aa86e09d0f390",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2618,null,null,2619,1094],"2bfa9270e51b94ada26b9541bd0c9260f0f46a1b03c7601844db3fe88308e967",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2620,null,null,2621,1097],"2ea84c7231f00ea9f8d616163690a33ca9b6c8e68c5f4498e616f1d57a583663",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2622,null,null,2623,1100],"ce5cd3e703de509c0f331808c88e9367babe2d84392f7111623c09d2f061a07a",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2624,null,null,2625,1091],"fe0572cef447f4b97e93a1199345fe11bebd2b0f016d087e40149d15d3422c00",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2626,null,null,2627,1094],"ec3c6bfbf4f61ccde3e22953a2b82ed6c85ca36d2f3013939730ec96c19f8020",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2628,null,null,2629,1097],"bacc60419538e9722eafc7d300ebc76c8a12cf432f3d5816aa2da1aea91e3d76",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2630,null,null,2631,1100],"d362e77df73535e661578293c8c4cf712510610acc0145fd1c8dda4d2e06400d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2632,null,null,2633,1091],"230b663d5264a9d46f5ebc753296ca226c2c51884c8a3ca548adce5c3ef52809",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2634,null,null,2635,1094],"68aa58001638b77805d1603097b1e68c0f842d7cf216ad3d488a92c70bfa2382",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2636,null,null,2637,1097],"6f2c3e5067bf9a0e0b2bb215726f5522de3f6ee98858744e96a1e9559ab7e0f6",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2638,null,null,2639,1100],"10f8f2862d1cc97ab8f02516dc9c699a53dc728edbbca23677a74abdc5efa211",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2640,null,null,2641,1091],"7c398875bbfb4151d349538a5c7e1bbdf53575137151ca93f2402bf35aa4f527",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2642,null,null,2643,1094],"7f516a6c34aa54298e7ea8d3f8daa3800b6ed48944112e35946fb6e11ef64a7d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2644,null,null,2645,1097],"7082be975c3ed5f46f854bdd467044b72d28990b51c7b27f0c3eddc83e5122df",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2646,null,null,2647,1100],"f6d3dcd54be3f2a8f8e50077b6ea1c35314321df900515fa45e286d0d53cc412",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16",0,0,"sm_120a","+ptx87","sm_120a",[2648,null,null,2649,1091],"f8f405a7c1b1473ec74ae7cdc80adfb8e19146fc65f0a3796c30584417acd47d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16",0,0,"sm_120f","+ptx88","sm_120f",[2650,null,null,2651,1094],"793d85ca5b1006842c8e3183a14fdda15979d00c9103b67ca527e8813184cd55",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16",0,0,"sm_121a","+ptx88","sm_121a",[2652,null,null,2653,1097],"f244fb98768b22a886d161eb9b9f00db1707104835cc8f2e7f2f5d8a2f03e12e",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16",0,0,"sm_121f","+ptx88","sm_121f",[2654,null,null,2655,1100],"a812a3c4389c5dd3a7263cfe69a2eae0400d5aa3cadb23674754670a579cde97",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2656,null,null,2657,1091],"4a87ecf6b081384eb33fa76e6bcda75564d9267bf2482f335b11e9a667df1a62",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2658,null,null,2659,1094],"78b64884f6dca3c93d8953892cfadb3cf7ca9b4a50c56eaa2c26554cd5a5e0c2",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2660,null,null,2661,1097],"5f2fc829907d8f8b4d36d07a3dc4c4693c79bee5ea10578f8ee8a98ddf2aa4e0",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2662,null,null,2663,1100],"c8546b20e75fb81657dfef52202850189339b9b2cd882c109c2c3f05067fa986",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2664,null,null,2665,1091],"87d8513032b4f67e1f24859cf4f1365ae40968d0992d911d35306a60280608c8",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2666,null,null,2667,1094],"a48569c490601c066de07f2caf194f1ba29eb93195ea1b707cb63f7a9d05911d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2668,null,null,2669,1097],"d83184492c09f5c00ef8cdfea13c3e910443a9270d7dceed1680b8f2642ee0aa",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2670,null,null,2671,1100],"bbb94ce42a916c27fb80e2124eac3403433f1286327d07018fc46c590a3d5ca4",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16",0,0,"sm_120a","+ptx87","sm_120a",[2672,null,null,2673,1091],"eb0dcdf37f61860cb132e65b7f79479478f63dd65703b0adc67e28c15a209d80",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16",0,0,"sm_120f","+ptx88","sm_120f",[2674,null,null,2675,1094],"862f0d5e2abd1e655a607765a7bbfd87694b63cb10d5030a2cd1e863176a1eae",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16",0,0,"sm_121a","+ptx88","sm_121a",[2676,null,null,2677,1097],"c5461030d1eb9b7d82c1c058001b3576c5dfdce67710b584dd8a1fec0799f5f4",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16",0,0,"sm_121f","+ptx88","sm_121f",[2678,null,null,2679,1100],"8d2974939758f3d6f1ad44167926be7cee8a0a2738b91b73815fbaed65d9dc1a",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16",0,0,"sm_120a","+ptx87","sm_120a",[2680,null,null,2681,1091],"eddf4805e9b40d9c592e703a79160114455e1448e878ae0a92c60e1023e0d119",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16",0,0,"sm_120f","+ptx88","sm_120f",[2682,null,null,2683,1094],"522eaaa58357ebb2bcd56393b0ed17ed8ec840866678e2c52e8305bec38e0bdc",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16",0,0,"sm_121a","+ptx88","sm_121a",[2684,null,null,2685,1097],"f10a7e80727f90ead77f689b0fea6cc14d18cdebec09153af8b2c6349962fbae",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16",0,0,"sm_121f","+ptx88","sm_121f",[2686,null,null,2687,1100],"8b05677f162342dfafef46eed3c3990417ab376a77404082ebb8ab25d9040f67",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32",0,0,"sm_120a","+ptx87","sm_120a",[2688,null,null,2689,1091],"cf865d84682a09d6dcf2eb7e905a78b0b2856c592611bbbc9516546dc2112dee",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2690,null,null,2691,1091],"bd008691a93a6ce34cf951fa83d3bb4f2ccc37baa72ffff83cb1d320da45ff67",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2692,null,null,2693,1091],"c6e86fc6a08604ead5cd69e352a2edbd43dcc27db0fd9a0eda92aa962fd66b38",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2694,null,null,2695,1091],"94288a75ead8cc13eea436c29f794eba65738e9218cb4f4f964c8452eae22ee9",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2696,null,null,2697,1091],"3246b73b545dbecd163dbaf232c386f393faf0dbe20428f82eb797e63700740d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32",0,0,"sm_120a","+ptx87","sm_120a",[2698,null,null,2699,1091],"7f1395a7acb5e222699616141e21cf3db0e75384b882bb2782e6549754b5baad",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2700,null,null,2701,1091],"3af97d6dec4d435a51fb2671bda8656459266c09cfebdbb55b5a04578812bf9d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2702,null,null,2703,1091],"2f218fed1a8244456886bdb6b78b690c34b59cd08b66de9f6a277c5483a5bc77",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2704,null,null,2705,1091],"7ff91c7ee153b20322ce10c4b407577d9d6f7453d42b480f2ebae70191014c84",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2706,null,null,2707,1091],"6a2170798a511b254fc03203b29b332ac54e599aadeb42172ca6caf05a023c1d",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32",0,0,"sm_120a","+ptx87","sm_120a",[2708,null,null,2709,1091],"82a398aad4bee18252629e6cb5ae29c48c43a46ed0b1aa4bc9b9dbbf58ed3427",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2710,null,null,2711,1091],"ff85bb8575cc17e46253604c23e82a607a30378054ee99066acc50ab3216aa1f",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2712,null,null,2713,1091],"4d2087cda6899bd4b4161d1ff99150f450e321d6341926c9575070d216b0a7fe",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2714,null,null,2715,1091],"00812ee1e0957679f89e191d53bb99d2efceeb4cc8bcd4f1c2eba7dadd2f7221",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2716,null,null,2717,1091],"2c211aae8291161f84ca0e5a2f507f4ce7c6cb8f7ad0b84fcd30d493cfbf9110",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32",0,0,"sm_120a","+ptx87","sm_120a",[2718,null,null,2719,1091],"a01f8120ea31ff2e9c5e9b48f13e7677b91424ea414ff236f045a3ed96ae3472",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2720,null,null,2721,1091],"90918f38302c4e3ea303c08d32539c1e813a3b6c6eb54096e93ff7b00a2967af",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2722,null,null,2723,1091],"bcb04e072f2452c9bceb6ca9896f3692d391d415db11b3d339d9dec6c731821b",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2724,null,null,2725,1091],"cf4641862ddcbdbc962a951a287153cbaff19dca3b5aaba37bc4cee88ab42aa3",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2726,null,null,2727,1091],"f507a2f9dae55a3eae7626c1ecb7002e2b316a6311e41d5e7f276186e92c5b02",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32",0,0,"sm_120a","+ptx87","sm_120a",[2728,null,null,2729,1091],"d1587da0fe4635a25c38117528064dd98bac77da98c5be69fffdf95f984d370e",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2730,null,null,2731,1091],"8ca202c6a29ad7b5947de043ef41603fb857cb44e4e2c7761d97c954f7c84663",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2732,null,null,2733,1091],"ee8f5d27a9d99903b912c20d9a1fea586da0385297c8eb4212fa29e040a598da",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32",0,0,"sm_120a","+ptx87","sm_120a",[2734,null,null,2735,1091],"92d46ba6a9c4452a92fddf06d891ab94817e06212d1056aa917836284232fbd6",0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32",0,0,"sm_120a","+ptx87","sm_120a",[2736,null,null,2737,1091],"889fe48fa0c1da3c0a0566b527e0b63f4914bf25d230ce1d640f167c8aef8527",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s4",0,0,"sm_80","+ptx85","sm_80",[2738,null,null,2739,2],"623a8984f421de974370c98f9592f3e69fcf9fc9d56567f4f8c4042872a9f5ad",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2740,null,null,2741,2],"98df013bddfb27d0ad45073d9355410b4cf57fbe93b556df13231f5fe34c1236",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s4_u4",0,0,"sm_80","+ptx85","sm_80",[2742,null,null,2743,2],"75ffac87a27fcd7b7aa2604d8234b4238570cfce7078d19733f0b82b947c2112",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2744,null,null,2745,2],"bd9a6eec94d9370a615ef15a4887c89282f728758aca9c9bc4865690ff0eb934",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s8",0,0,"sm_80","+ptx85","sm_80",[2746,null,null,2747,2],"848275afdaf91665228cd422fc24022862ba6cd0afc2ce938bdd7eabd4a26c61",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2748,null,null,2749,2],"0818575a5904fb142d1f365f9f5ca7d9c6a1a3a311ff7807b91d7e3a77a0bcde",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s8_u8",0,0,"sm_80","+ptx85","sm_80",[2750,null,null,2751,2],"0f09b739ef0c16aa6771bdb8e2de77125d580081f100fab5b58ebfb8ea367838",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2752,null,null,2753,2],"797f27f7ce242cc82586c92330cb27e2505055f898889d618d240bfd43992c73",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u4",0,0,"sm_80","+ptx85","sm_80",[2754,null,null,2755,2],"e5124ccbb70317a2bc335a5a7ec2feabb5fd42b4a9412b3bf96586bf0ca5c477",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u4_s4",0,0,"sm_80","+ptx85","sm_80",[2756,null,null,2757,2],"7c2e28aae97dceceff93f67c3c4b6acb5449d4cff645fcd6220465735940ad9b",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2758,null,null,2759,2],"c2d02ac2640a1a4ff4ae20c4fe9ed5a765ee034659acf310c6072dbfe78f190f",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite",0,0,"sm_80","+ptx85","sm_80",[2760,null,null,2761,2],"196b855e9979fb29efe4a354c8d413a7bba81c5e02e23757047f5f4feb5ef99e",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u8",0,0,"sm_80","+ptx85","sm_80",[2762,null,null,2763,2],"3b8738ae3924ced53ac1015453fb029b021712e3d49f9d8c5517bf01266d8034",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u8_s8",0,0,"sm_80","+ptx85","sm_80",[2764,null,null,2765,2],"1da7e00a5518eb689cf08011ae123d23e5dce11a7af8689fc2dfc94407a2483a",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2766,null,null,2767,2],"ff673243770c2b4b2fe0731cf70c2c09ad4a278d38d856c2f792c8011ba4720a",0]
+["mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite",0,0,"sm_80","+ptx85","sm_80",[2768,null,null,2769,2],"127b05cf0d50517df21d7fb6f840a9fe4c342e22a4441a94924f34b05ed7b94d",0]
+["mma_sp_ordered_metadata_m16n8k8_f32_tf32",0,0,"sm_80","+ptx85","sm_80",[2770,null,null,2771,2],"2515876dfbb516202f0a61fbeec56c9c52b5f8fe471694ea931692c672b65b54",0]
+["movmatrix_trans_b16",0,0,"sm_75","+ptx78","sm_75",[2772,null,null,2773,7],"f977fb79e1cc5a0394af583272f3850fc0dd7ee2a542c3197eb193aad5f39bd3",0]
+["mul_bf16x2",0,0,"sm_90","+ptx78","sm_90",[2774,null,null,2775,14],"834618892fcacf27b71218424840868388ad296c94056a6289a691aed82d2206",0]
+["mul_f16x2",0,0,"sm_70","+ptx60",null,[2776,null,null,2777,null],"f4c90564e551287271cdb171b63b2f649e78ce806891e5fc1a5eafd0bb1a42bb",0]
+["mul_f16x2",1,0,"sm_70","+ptx60","sm_75",[2778,null,null,2777,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["mul_f32x2",0,0,"sm_100","+ptx86","sm_100",[2779,null,null,2780,21],"e30ba35077b9fabc66d1c82f71edd065adb9306575a4c23e176ec5dc7ba79cc2",0]
+["mul_ftz_f32x2",0,0,"sm_100","+ptx86","sm_100",[2781,null,null,2782,21],"1cd3f4a46861fb83b4062a3a75d424cd30cfb155a1aeff0b7c50cdfa6b2c5845",0]
+["mul_rm_f32",1,0,"sm_80","+ptx70","sm_80",[2783,2786,2784,2785,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rm_f64",1,0,"sm_80","+ptx70","sm_80",[2787,2790,2788,2789,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rm_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2791,2794,2792,2793,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rn_f32",1,0,"sm_80","+ptx70","sm_80",[2795,2798,2796,2797,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rn_f64",1,0,"sm_80","+ptx70","sm_80",[2799,2802,2800,2801,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rn_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2803,2806,2804,2805,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rp_f32",1,0,"sm_80","+ptx70","sm_80",[2807,2810,2808,2809,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rp_f64",1,0,"sm_80","+ptx70","sm_80",[2811,2814,2812,2813,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rp_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2815,2818,2816,2817,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rz_f32",1,0,"sm_80","+ptx70","sm_80",[2819,2822,2820,2821,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rz_f64",1,0,"sm_80","+ptx70","sm_80",[2823,2826,2824,2825,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["mul_rz_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2827,2830,2828,2829,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["nanosleep",0,0,"sm_70","+ptx63",null,[2831,null,null,2832,null],"e6b35e91f3eb1fb4fdf4f848358073d177bd079a7521623f63c5df3b1593a9fd",0]
+["nanosleep",0,0,"sm_75","+ptx63","sm_75",[2833,null,null,2834,7],"5c3e47fd18cde1ac5d1d3c98376cb5fefbbd2620453e4f2f8d4a510422b3cf75",0]
+["neg_bf16x2",0,0,"sm_80","+ptx70","sm_80",[2835,null,null,2836,2],"a77bb217c46e9f4cc45b02bd3e2e7f43fb021716ad20c0fe6c5083a50abd7447",0]
+["neg_f16x2",0,0,"sm_70","+ptx60",null,[2837,null,null,2838,null],"66b86ece7e9bd16687c799400c86c23d7e86b0e5bf246567293a3cb674e52d04",0]
+["neg_f16x2",1,0,"sm_70","+ptx60","sm_75",[2839,null,null,2838,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["nsmid",1,0,"sm_20","+ptx32",null,[2840,2843,2841,2842,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["nsmid",1,0,"sm_20","+ptx32","sm_75",[2840,2843,2841,2842,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["nwarpid",1,0,"sm_20","+ptx32",null,[2844,2847,2845,2846,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["nwarpid",1,0,"sm_20","+ptx32","sm_75",[2844,2847,2845,2846,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["packed_atomic_add_bf16x2",0,0,"sm_90","+ptx78","sm_90",[2848,null,null,2849,14],"1fc329cc73b40ee07454dc4d097e5a1c0e811b34b11407dfd9d9b3a1abf83a2d",0]
+["packed_atomic_add_f16x2",0,0,"sm_70","+ptx62",null,[2850,null,null,2851,null],"6cc9c0a97fb7b2a69722f12d763a01d096b007ca488b0a5a18b4683002799a2d",0]
+["packed_atomic_add_f16x2",1,0,"sm_70","+ptx62","sm_75",[2852,null,null,2851,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["pmevent",0,0,"sm_20","+ptx32",null,[2853,null,null,2854,null],"b31291b4d9fc4b3b71f480bf0c30bb03a0d9780dbb9b31f197bcb3fee022f9e8",0]
+["pmevent",1,0,"sm_20","+ptx32","sm_75",[2855,null,null,2854,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prefetch_tma_descriptor",0,0,"sm_90","+ptx80","sm_90",[2856,2857,2858,2859,14],"bd75db27d12c32c18ca977d70274cb8a5d495584b761de38a57d326e1799b223",0]
+["prmt",0,0,"sm_20","+ptx32",null,[2860,2861,2862,2863,null],"9038cd62eb938b71c62abcff593ba93d392560fb3961cefff55bb0b13451e5ce",0]
+["prmt",1,0,"sm_20","+ptx32","sm_75",[2864,2866,2865,2863,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prmt_b4e",0,0,"sm_20","+ptx32",null,[2867,2868,2869,2870,null],"39a39acad916c5b4f114998541446ebf46f8761832d6475af67ad70dd29f91c2",0]
+["prmt_b4e",1,0,"sm_20","+ptx32","sm_75",[2871,2873,2872,2870,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prmt_ecl",0,0,"sm_20","+ptx32",null,[2874,2875,2876,2877,null],"43a89345cdd683ee3a3387246cf0c9d581abd4f6f9f89f20df6cd8008aacce94",0]
+["prmt_ecl",1,0,"sm_20","+ptx32","sm_75",[2878,2880,2879,2877,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prmt_ecr",0,0,"sm_20","+ptx32",null,[2881,2882,2883,2884,null],"b47372f2d609755c4d5074c5de5946163f7a582de10edb938eb2798cf0acb41c",0]
+["prmt_ecr",1,0,"sm_20","+ptx32","sm_75",[2885,2887,2886,2884,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prmt_f4e",0,0,"sm_20","+ptx32",null,[2888,2889,2890,2891,null],"6557d45b6c4806f5c6ccdebd29028aec41aad8f6cfe52677f33c3dc4829d3fa1",0]
+["prmt_f4e",1,0,"sm_20","+ptx32","sm_75",[2892,2894,2893,2891,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prmt_rc16",0,0,"sm_20","+ptx32",null,[2895,2896,2897,2898,null],"00123086e52adbadf9e36254ef5e2521bc8db523d4563754ff68789221fac0e0",0]
+["prmt_rc16",1,0,"sm_20","+ptx32","sm_75",[2899,2901,2900,2898,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["prmt_rc8",0,0,"sm_20","+ptx32",null,[2902,2903,2904,2905,null],"56cb3427c654f71c57131c195824645ae24aade10d0a406f703b79491b4288b9",0]
+["prmt_rc8",1,0,"sm_20","+ptx32","sm_75",[2906,2908,2907,2905,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["rcp_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2909,2912,2910,2911,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_approx_ftz_f64",1,0,"sm_80","+ptx70","sm_80",[2913,2916,2914,2915,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rm_f32",1,0,"sm_80","+ptx70","sm_80",[2917,2920,2918,2919,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rm_f64",1,0,"sm_80","+ptx70","sm_80",[2921,2924,2922,2923,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rm_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2925,2928,2926,2927,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rn_f32",1,0,"sm_80","+ptx70","sm_80",[2929,2932,2930,2931,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rn_f64",1,0,"sm_80","+ptx70","sm_80",[2933,2936,2934,2935,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rn_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2937,2940,2938,2939,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rp_f32",1,0,"sm_80","+ptx70","sm_80",[2941,2944,2942,2943,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rp_f64",1,0,"sm_80","+ptx70","sm_80",[2945,2948,2946,2947,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rp_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2949,2952,2950,2951,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rz_f32",1,0,"sm_80","+ptx70","sm_80",[2953,2956,2954,2955,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rz_f64",1,0,"sm_80","+ptx70","sm_80",[2957,2960,2958,2959,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rcp_rz_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[2961,2964,2962,2963,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["redux_sync_add",0,0,"sm_80","+ptx70",null,[2965,2966,2967,2968,null],"e80272156893da46ca0de4eddbe6fd1d1e21a20e020087cbae53f545ac7bcd3d",0]
+["redux_sync_and",0,0,"sm_80","+ptx70",null,[2969,2970,2971,2972,null],"dd6dff86c8dc071d4a1822a676a8747aeaa9b80c1c7944d9c942d7e984dddd35",0]
+["redux_sync_max_abs_f32",0,0,"sm_100a","+ptx86",null,[2973,2974,2975,2976,null],"e28a09bc2a75b1d56efeae4440efa6e9b7f0494374fb9e1555caefdff1e3942e",0]
+["redux_sync_max_abs_nan_f32",0,0,"sm_100a","+ptx86",null,[2977,2978,2979,2980,null],"4eb40e97df368372b8681148f8c867e0732012c19ed88b9ca97ea9936570d6ae",0]
+["redux_sync_max_f32",0,0,"sm_100a","+ptx86",null,[2981,2982,2983,2984,null],"575070cf759091b21b81fb520d46afde0a6f2c1924f345e85d289b0811cb61d1",0]
+["redux_sync_max_i32",0,0,"sm_80","+ptx70",null,[2985,2986,2987,2988,null],"f0833bc2e968c85b757e090bde4b6c3c3ab5a7c3cdc8cd51b89df47e9c34137d",0]
+["redux_sync_max_nan_f32",0,0,"sm_100a","+ptx86",null,[2989,2990,2991,2992,null],"3e368d0ca892e87d27000de2b93e0db53f4f779d259109aed0a4af993f3bc7bc",0]
+["redux_sync_max_u32",0,0,"sm_80","+ptx70",null,[2993,2994,2995,2996,null],"93ef76dd6c2eccf2b94306808f317c0c65fc0b5f6b9563a26423e1293e78e8c0",0]
+["redux_sync_min_abs_f32",0,0,"sm_100a","+ptx86",null,[2997,2998,2999,3000,null],"c336e9b136d922a81f8d9046b1517c495b37e17d54772a7a7cd2ae943575fe43",0]
+["redux_sync_min_abs_nan_f32",0,0,"sm_100a","+ptx86",null,[3001,3002,3003,3004,null],"095f7590e7174f17f49476fc42b192a682a15a9e006183b7e2ef4d5e69f6c913",0]
+["redux_sync_min_f32",0,0,"sm_100a","+ptx86",null,[3005,3006,3007,3008,null],"cdb063db08cd6048a67b9dcf4f13a06928e222fd97300d33acf8339759a43453",0]
+["redux_sync_min_i32",0,0,"sm_80","+ptx70",null,[3009,3010,3011,3012,null],"ffae2eed4da424044dfaa7b83f53fa58cffcacce719f3cc09104d35cb47e4e73",0]
+["redux_sync_min_nan_f32",0,0,"sm_100a","+ptx86",null,[3013,3014,3015,3016,null],"1cda15ce68b50dd1d7963a0073ecaaf9570ec2be82be9b64afbff8548e83bd8c",0]
+["redux_sync_min_u32",0,0,"sm_80","+ptx70",null,[3017,3018,3019,3020,null],"fce59e9afc230279936b8d16b5169285a1cde494458877f7cad82f7b8c96bfef",0]
+["redux_sync_or",0,0,"sm_80","+ptx70",null,[3021,3022,3023,3024,null],"a15a0bdbc7980416d6d13c59cba7ad7428bd0ba02dfc8d7a1507f65ae9c68c9c",0]
+["redux_sync_xor",0,0,"sm_80","+ptx70",null,[3025,3026,3027,3028,null],"ae9477d883c79819241d05a121ee3a2fb3ab776bf35254c7c00e592010c152d2",0]
+["rsqrt_approx_f32",1,0,"sm_80","+ptx70","sm_80",[3029,null,null,3030,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rsqrt_approx_f64",1,0,"sm_80","+ptx70","sm_80",[3031,null,null,3032,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["rsqrt_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3033,null,null,3034,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["setmaxnreg_dec",0,0,"sm_90a","+ptx80","sm_90a",[3035,3036,3037,3038,3039],"19d9a310aae3fa355f4b539c6499fd5984890a1f7a6529e808f174319f5deaa9",0]
+["setmaxnreg_inc",0,0,"sm_90a","+ptx80","sm_90a",[3040,3041,3042,3043,3039],"a6be87eb707e8ee05c6a358549b548c8d655d32225ccaff5e30c5d11515d258c",0]
+["shuffle_down_f32_sync",0,0,"sm_30","+ptx60",null,[3044,3045,3046,3047,null],"290a478e594ee1d850cbf541220031cffbca3926ab3f1785d3a4415363754264",0]
+["shuffle_down_f32_sync",1,0,"sm_30","+ptx60","sm_75",[3048,3050,3049,3047,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_down_sync",0,0,"sm_30","+ptx60",null,[3051,3052,3053,3054,null],"d40d725fa7de2b930c3500d77047ca5ffafd8543b8639887dc737326e9506900",0]
+["shuffle_down_sync",1,0,"sm_30","+ptx60","sm_75",[3055,3057,3056,3054,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_down_u64_sync",0,0,"sm_30","+ptx60",null,[3058,null,null,3059,null],"11938deac5893f8108008170486a0b2bf960c34ed3026ae212b62ce689bb2738",0]
+["shuffle_down_u64_sync",1,0,"sm_30","+ptx60","sm_75",[3060,null,null,3059,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_f32_sync",0,0,"sm_30","+ptx60",null,[3061,3062,3063,3064,null],"600d41978aaa066ee43dec361554feae2d8ca4769c1ed8f98629a450a13a096d",0]
+["shuffle_f32_sync",1,0,"sm_30","+ptx60","sm_75",[3065,3067,3066,3064,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_sync",0,0,"sm_30","+ptx60",null,[3068,3069,3070,3071,null],"e437be465942c42f1b4ac78db2150dff97bef7178baa435251ee3f5c2cc595db",0]
+["shuffle_sync",1,0,"sm_30","+ptx60","sm_75",[3072,3074,3073,3071,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_u64_sync",0,0,"sm_30","+ptx60",null,[3075,null,null,3076,null],"f8c996799f8754fd27e99ed503dd27a75d31ece3114757974a9fa8905a9d423b",0]
+["shuffle_u64_sync",1,0,"sm_30","+ptx60","sm_75",[3077,null,null,3076,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_up_f32_sync",0,0,"sm_30","+ptx60",null,[3078,3079,3080,3081,null],"8163fb66de75d2f64343c94a68b07a132b639c13588b6f1332fdd3af30ce32ad",0]
+["shuffle_up_f32_sync",1,0,"sm_30","+ptx60","sm_75",[3082,3084,3083,3081,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_up_sync",0,0,"sm_30","+ptx60",null,[3085,3086,3087,3088,null],"7fdf9a0e0ac2e04015c20114657e4b74bea13e6f610c92b999f8bfaf0135784e",0]
+["shuffle_up_sync",1,0,"sm_30","+ptx60","sm_75",[3089,3091,3090,3088,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_up_u64_sync",0,0,"sm_30","+ptx60",null,[3092,null,null,3093,null],"5969f4f2f726b8c138d5a1831f54db05a6243b016a8c4c40039e35ba00ae4a89",0]
+["shuffle_up_u64_sync",1,0,"sm_30","+ptx60","sm_75",[3094,null,null,3093,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_xor_f32_sync",0,0,"sm_30","+ptx60",null,[3095,3096,3097,3098,null],"5455d57b67bb03e1efe76c91a355cfe6f23a7f5b27dc130198de11ab9c74579e",0]
+["shuffle_xor_f32_sync",1,0,"sm_30","+ptx60","sm_75",[3099,3101,3100,3098,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_xor_sync",0,0,"sm_30","+ptx60",null,[3102,3103,3104,3105,null],"5a1c74e474bfc9bdbdff3c8c4e01f40961d7aad96003cd78ed13ac6d35902e18",0]
+["shuffle_xor_sync",1,0,"sm_30","+ptx60","sm_75",[3106,3108,3107,3105,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["shuffle_xor_u64_sync",0,0,"sm_30","+ptx60",null,[3109,null,null,3110,null],"1a9d47c194c2dc3caa70d9491988960b1891733bd60f081b7f0789800d2aa866",0]
+["shuffle_xor_u64_sync",1,0,"sm_30","+ptx60","sm_75",[3111,null,null,3110,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["sin_approx_f32",1,0,"sm_80","+ptx70","sm_80",[3112,null,null,3113,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sin_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3114,null,null,3115,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["smid",1,0,"sm_20","+ptx32",null,[3116,null,null,3117,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["smid",1,0,"sm_20","+ptx32","sm_75",[3116,null,null,3117,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_approx_f32",1,0,"sm_80","+ptx70","sm_80",[3118,3121,3119,3120,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_approx_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3122,3125,3123,3124,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rm_f32",1,0,"sm_80","+ptx70","sm_80",[3126,3129,3127,3128,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rm_f64",1,0,"sm_80","+ptx70","sm_80",[3130,3133,3131,3132,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rm_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3134,3137,3135,3136,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rn_f32",1,0,"sm_80","+ptx70","sm_80",[3138,3141,3139,3140,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rn_f64",1,0,"sm_80","+ptx70","sm_80",[3142,3145,3143,3144,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rn_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3146,3149,3147,3148,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rp_f32",1,0,"sm_80","+ptx70","sm_80",[3150,3153,3151,3152,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rp_f64",1,0,"sm_80","+ptx70","sm_80",[3154,3157,3155,3156,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rp_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3158,3161,3159,3160,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rz_f32",1,0,"sm_80","+ptx70","sm_80",[3162,3165,3163,3164,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rz_f64",1,0,"sm_80","+ptx70","sm_80",[3166,3169,3167,3168,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["sqrt_rz_ftz_f32",1,0,"sm_80","+ptx70","sm_80",[3170,3173,3171,3172,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["stmatrix_m8n8_x2_b16",0,0,"sm_90","+ptx78","sm_90",[3174,3175,3176,3177,14],"bc7b37aa1b0195ff2a1a5743200e560b9e0a84321a784f98b3126a0cbe268406",0]
+["stmatrix_m8n8_x2_trans_b16",0,0,"sm_90","+ptx78","sm_90",[3178,3179,3180,3181,14],"e258409828b040ea90d4b3b9b6f0dcd1a23902a9dfa17bc640c7d3db9e118018",0]
+["stmatrix_m8n8_x4_b16",0,0,"sm_90","+ptx78","sm_90",[3182,3183,3184,3185,14],"0a44eaf197e9bc96929b824e4dfe27cf4473c4f4780951463b4ed108d084d50c",0]
+["stmatrix_m8n8_x4_trans_b16",0,0,"sm_90","+ptx78","sm_90",[3186,3187,3188,3189,14],"70d137cee37b3162904c822964695c12e294239d8dc3e070dc95338dbbdfa3a9",0]
+["sub_bf16x2",0,0,"sm_90","+ptx78","sm_90",[3190,null,null,3191,14],"808ee0771bce0c35e6ff61009332bdaaf9e4f609bcfe6255a53265dfdfb09d4f",0]
+["sub_f16x2",0,0,"sm_70","+ptx60",null,[3192,null,null,3193,null],"9eed8aec0a7b046a2a2d74cf78507fae943bdd3f6e26a15975bd5d9df57bdcad",0]
+["sub_f16x2",1,0,"sm_70","+ptx60","sm_75",[3194,null,null,3193,18],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["sub_f32x2",0,0,"sm_100","+ptx86","sm_100",[3195,null,null,3196,21],"2c58c432f75e20d88db1ab13ccd09bec61622aa9333e4c0aa3da8a9dea47a3e1",0]
+["sub_ftz_f32x2",0,0,"sm_100","+ptx86","sm_100",[3197,null,null,3198,21],"858219cf70a211e7804a883780a614697829385626beebe9f699fe7a488e5cf3",0]
+["sync_mask",0,0,"sm_30","+ptx60",null,[3199,3200,3201,3202,null],"8576497798ed0743dda45f25676dda80073e50e9bb70d0a53108249c91189678",0]
+["sync_mask",1,0,"sm_30","+ptx60","sm_75",[3203,3205,3204,3202,138],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["sync_threads",0,0,"sm_20","+ptx32",null,[3206,3207,3208,3209,null],"7ef2a8041f97ddd2998ae613216801fd4b27ac9dadcf7ddb5205ef704fe321d6",0]
+["sync_threads",1,0,"sm_20","+ptx32","sm_75",[3210,3212,3211,3209,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["tanh_approx_f32",1,0,"sm_80","+ptx70","sm_80",[3213,null,null,3214,2],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["tcgen05_alloc",0,0,"sm_100a","+ptx86","sm_100a",[3215,null,null,3216,3217],"58fe9a03e4e11fe4d036a9fafcfdbc47e335d36ed9f7eb17649a598c1be3cf3c",0]
+["tcgen05_alloc_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3218,null,null,3219,3217],"ec8824469a25d273a7a49acde5c6130429e03115dc4de2ebfd76309ad0a46aa5",0]
+["tcgen05_commit",0,0,"sm_100a","+ptx86","sm_100a",[3220,null,null,3221,213],"f3a87cd4c378ca65c4942cf248db22784450236c315c108fda49d410cbd72190",0]
+["tcgen05_commit_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3222,null,null,3223,213],"b061f05aa0880c1492b5ce3815b012677bc30efb50c3bc4b1bac5aecea41573e",0]
+["tcgen05_commit_multicast",0,0,"sm_100a","+ptx86","sm_100a",[3224,null,null,3225,213],"9ac788e4c4310d8392217f96ed668addfdc158daab78bf07898179a429e8b0d8",0]
+["tcgen05_commit_multicast_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3226,null,null,3227,213],"d90901cf62977774fe658d128638dba4d600ee977cb86a54e966491512a43a72",0]
+["tcgen05_commit_shared_cluster",0,0,"sm_100a","+ptx86","sm_100a",[3228,null,null,3229,213],"ce95d30de242938d29948f249a6518afb960751d5d8954dd083a2f560acb4103",0]
+["tcgen05_commit_shared_cluster_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3230,null,null,3231,213],"e320f11140c36e3cbbe1bda8763815d7780645617dbb308a062203e8727146b5",0]
+["tcgen05_cp_128x128b",0,0,"sm_100a","+ptx86","sm_100a",[3232,null,null,3233,213],"72b3b5c05f1b4cacbdbe30c438ef97d0e200c8476929fdb5929cb7b0e6440139",0]
+["tcgen05_cp_128x128b_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[3234,null,null,3235,213],"b84e07d047276ee1ed11d5cb8126d6a84dcf222d1ccba52257901c77baf0edf0",0]
+["tcgen05_cp_128x128b_b4x16_p64_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3236,null,null,3237,213],"8e87c580ab5e7cff51de4609159b2963bece1cd08190e6d364f172e94ebfb8eb",0]
+["tcgen05_cp_128x128b_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[3238,null,null,3239,213],"5d944a497494c4903b3f3856e2e88523a130fad0b5350d43bda81d354e98d4c7",0]
+["tcgen05_cp_128x128b_b6x16_p32_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3240,null,null,3241,213],"1de4911d64cd44fdea49c216b89d3dda070448ae4eb1370ab42dfde9b5ef57fe",0]
+["tcgen05_cp_128x128b_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3242,null,null,3243,213],"0ad113d14abce959ebb368ed4a69570b24a49daf043c721101c289dc2067ce55",0]
+["tcgen05_cp_128x256b_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[3244,null,null,3245,213],"88b37501210a51bf95b090ae5b4b19d596c70add937c4801eac372c06880acbe",0]
+["tcgen05_cp_128x256b_b4x16_p64_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3246,null,null,3247,213],"cd764387803214f7e95dbeb31998f52a9343f98f613764341895801c3f494a3e",0]
+["tcgen05_cp_128x256b_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[3248,null,null,3249,213],"81a6a70127aade8f6815843003fd5b9fc1a507caf2765a45a22596b5d8c14e5c",0]
+["tcgen05_cp_128x256b_b6x16_p32_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3250,null,null,3251,213],"b38c3e07e8f0f2ca98f19887ba8529d2e8ff90cb19057015de91762e01015795",0]
+["tcgen05_cp_32x128b_warpx4",0,0,"sm_100a","+ptx86","sm_100a",[3252,null,null,3253,213],"bb4ce0af69ed85df783f6ab76a6ae97645677177b1d44170315d9052dbb3faa0",0]
+["tcgen05_cp_32x128b_warpx4_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[3254,null,null,3255,213],"78c024d318b6a20f6469e3a18c0cbf9638567689681f6cbdfa332dacfe2a5ec4",0]
+["tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3256,null,null,3257,213],"97000dcbf70a3ec8afae5ff3be5c3cfecf5aa74e73f44b1dd13f510977fbb7d9",0]
+["tcgen05_cp_32x128b_warpx4_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[3258,null,null,3259,213],"fa22b999480c289272df9e7e929896a0dd2ffb7c645a448f87d987ca41c2510a",0]
+["tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3260,null,null,3261,213],"6a9bf6e439806c3cedcf92dd114d43a8ce31e84b1287f57dec593b4bb971c360",0]
+["tcgen05_cp_32x128b_warpx4_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3262,null,null,3263,213],"7c5ab606065394d101ce3df1c64f7631f4d60d710171cc2eb9782cfa0cae4fe9",0]
+["tcgen05_cp_4x256b",0,0,"sm_100a","+ptx86","sm_100a",[3264,null,null,3265,213],"ef5e815759387960cf159e792ce6c953dad75e445ecf410562ca584bd5bf5aaf",0]
+["tcgen05_cp_4x256b_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[3266,null,null,3267,213],"0f57329dd6a23aa75a7072e42ea3982a1fadd2b5f0fde4baace48ca41b883d53",0]
+["tcgen05_cp_4x256b_b4x16_p64_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3268,null,null,3269,213],"fd580540cbe0e62e6ddb926732fa6c9e9286f2c285ec4206d0d7458b559d32f7",0]
+["tcgen05_cp_4x256b_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[3270,null,null,3271,213],"5dda004e258505786706dfbabe2ccff0107a1a9ebfd3a7787963cc67e285b2e4",0]
+["tcgen05_cp_4x256b_b6x16_p32_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3272,null,null,3273,213],"a84a8af23da893b0f656f5fd5af916a2383c718ab6d98049fa6b21308602d693",0]
+["tcgen05_cp_4x256b_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3274,null,null,3275,213],"5fb68fb2377cf53502147ecc817f6843a9f80c597279253036e5af7c065ee741",0]
+["tcgen05_cp_64x128b_warpx2_01_23",0,0,"sm_100a","+ptx86","sm_100a",[3276,null,null,3277,213],"c0543d0245756d91e8b231f6935ab1da127d2900c9a21fe861f737542a43171c",0]
+["tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[3278,null,null,3279,213],"1df0477a2a18c301a4cc0d583a4aa42ebb3abe2be5bcf0ee03ed28e67659320c",0]
+["tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3280,null,null,3281,213],"62245fe5b23deff18e4a6f0c05b66f737f942b84d5e7b5abe53c054f42a8b31a",0]
+["tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[3282,null,null,3283,213],"f43a32c767fd0f58116000277b943ed94a0783690dcbb2076afc74984c4eaeb4",0]
+["tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3284,null,null,3285,213],"222361e34d5691ce3b787850aed5d55a9982096751488342ed89c448d06330e2",0]
+["tcgen05_cp_64x128b_warpx2_01_23_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3286,null,null,3287,213],"0df30c0115af8bb9709b9a5acdc05d69b1952554f794b198d96da1bb32195c1f",0]
+["tcgen05_cp_64x128b_warpx2_02_13",0,0,"sm_100a","+ptx86","sm_100a",[3288,null,null,3289,213],"94af839019dfe216ce9f1a311ed5cfce286adf952c61075f22a7ddb96ef503ae",0]
+["tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64",0,0,"sm_100a","+ptx86","sm_100a",[3290,null,null,3291,213],"250601e68bf1515bf61fb57db1c04f979dffc8f1b197c21437c10a7daaf96910",0]
+["tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3292,null,null,3293,213],"ba8c16913f19518a4cb6093b3374339128262bc93b957a26abf56cb670117c31",0]
+["tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32",0,0,"sm_100a","+ptx86","sm_100a",[3294,null,null,3295,213],"3025814f4c73c21da65d2e7ddc71f3d6de8b627515ad2fc6e35f61a6edf2897a",0]
+["tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3296,null,null,3297,213],"d58f901a2d66febbb11f4ecbfa41872d94d87f8a29536fab485e518cf354351b",0]
+["tcgen05_cp_64x128b_warpx2_02_13_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3298,null,null,3299,213],"4174ca47cbd41f97b96a2bcf1ac64c64b446de7b5586a76a8bf0480bb15b63cb",0]
+["tcgen05_cp_smem_to_tmem",0,0,"sm_100a","+ptx86","sm_100a",[3300,null,null,3301,213],"f1999f11aaf8ea91ee3acc6390f605225b0c1e0fde2b4faf89139b923b158643",0]
+["tcgen05_cp_smem_to_tmem_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3302,null,null,3303,213],"0112fe9595a57445ea717958f4100ed17bfecb31ac0ba9fbdf40bedfdafdff5d",0]
+["tcgen05_dealloc",0,0,"sm_100a","+ptx86","sm_100a",[3304,null,null,3305,3217],"965b54e1e94057a2e0aed3065659c0b72eb42967f601e7efe1be9dc570846756",0]
+["tcgen05_dealloc_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3306,null,null,3307,3217],"17898ad0d3a6a124e581d21850b4dac70ad391567bba43cd5f67bb607d7c0519",0]
+["tcgen05_fence_after_thread_sync",0,0,"sm_100a","+ptx86","sm_100a",[3308,null,null,3309,213],"85e02704b38697b96d2e6fd0c7d295f464966e7f9327404d9e2dc1f8828bc189",0]
+["tcgen05_fence_before_thread_sync",0,0,"sm_100a","+ptx86","sm_100a",[3310,null,null,3311,213],"e03db9be10cc25ee9ffb7d3b576796e170294ffdc4a151186c3f78a1519be991",0]
+["tcgen05_ld_16x128b_x16_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3312,null,null,3313,213],"4d99cbed261a05d2ab4ace6c12bd385b07ae54efa95036b69383c016b6261b75",0]
+["tcgen05_ld_16x128b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3314,null,null,3315,213],"a0eab2ef44994ddc1b25d58886888c4a7ee5feb4297de4404007173e60ace550",0]
+["tcgen05_ld_16x128b_x1_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3316,null,null,3317,213],"51bf1a67352c2caa502bcb9109e2134522b79848c45b5d2ab2da82ee0a7e9cb8",0]
+["tcgen05_ld_16x128b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3318,null,null,3319,213],"e23632ed6d23c07048f215c01b04ad3118ea60503d712145dbe88bb2e9a190b4",0]
+["tcgen05_ld_16x128b_x2_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3320,null,null,3321,213],"2dda61b4bdc2b87ff92fb40153fe9644680bfb1ca94f30c04552b1f90e0c40a7",0]
+["tcgen05_ld_16x128b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3322,null,null,3323,213],"4a20141ed7c740878f471ce7e2bbaf0714ba4de67bf2b96c5fb2c8788a7aa6ad",0]
+["tcgen05_ld_16x128b_x32_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3324,null,null,3325,213],"6cd835572c3240a488e3dc74b469b98411270a37b4616d2480685acb628b78cc",0]
+["tcgen05_ld_16x128b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3326,null,null,3327,213],"2000d4154ba0cf3670a131096deb8c0b06464c1fc7f7db190969b8d89df07714",0]
+["tcgen05_ld_16x128b_x4_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3328,null,null,3329,213],"d17d1dd7edc8ce8b647dfb7461bb0de0fc98c604bdc17ff6854a5d0266581ad3",0]
+["tcgen05_ld_16x128b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3330,null,null,3331,213],"01f352b287ffe8b98cdc558c34d3f0871979d3c94aa39e130a9a248302a6eee8",0]
+["tcgen05_ld_16x128b_x64_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3332,null,null,3333,213],"af210a5cd3bbc232a60c3fe242ef47d76218e58dbbfaf73c3752b3c29efeea71",0]
+["tcgen05_ld_16x128b_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3334,null,null,3335,213],"1ba118e846b68061595bbd789591d5f27b5ff20ea26994a636f683d3717d5603",0]
+["tcgen05_ld_16x128b_x8_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3336,null,null,3337,213],"4f211b3114711221ec71c84a83cf8a9ebef91139a09906357ed468441f5f4a81",0]
+["tcgen05_ld_16x128b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3338,null,null,3339,213],"24598697c98b9cdacea2b8b088d4e2a507c8281c83738409856d3e836faa7ed8",0]
+["tcgen05_ld_16x256b_pure",0,0,"sm_100a","+ptx86","sm_100a",[3340,null,null,3341,213],"402119b41e458ae87d537356949636ede4799c07423e4247c6832d0fbee61448",0]
+["tcgen05_ld_16x256b_x16_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3342,null,null,3343,213],"914c23ca33289337cbf8d705674485fcb80140dad51bee748484b12b55b2381f",0]
+["tcgen05_ld_16x256b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3344,null,null,3345,213],"47ae1797b9a1c5cddfdc8f5134dcb535623ac1e1930cdb54596af4be62fcb0e0",0]
+["tcgen05_ld_16x256b_x1_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3346,null,null,3347,213],"b72573f8134f598dd9bbe1894900a54a3e2944e25da1b7791b61da7d03ac6f86",0]
+["tcgen05_ld_16x256b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3348,null,null,3349,213],"5c53a867333c1f3be313f60ca60e17d0faa0533b017ae168e443cb6ff759b201",0]
+["tcgen05_ld_16x256b_x2_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3350,null,null,3351,213],"f67af149f9e5cad987ade6d81b9d185b2bff5752e309be97562414e5f107d60e",0]
+["tcgen05_ld_16x256b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3352,null,null,3353,213],"e867fab3e6b128f5c3caeb860132d30a086f71200830f13cee09cc4064aaa9e6",0]
+["tcgen05_ld_16x256b_x32_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3354,null,null,3355,213],"03d213c56ae57abf96d96fd40c3d090120b3c5e7a052306919b3882445eb1722",0]
+["tcgen05_ld_16x256b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3356,null,null,3357,213],"80ff67d40f73fe4ccd6b32fb7b6cf70a7241ab227a6d829e224c859a97540ae0",0]
+["tcgen05_ld_16x256b_x4_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3358,null,null,3359,213],"78cd435c3cae8c775c4d27ea154c1ddef9d02afe175b57f8752d3ac0c2669c24",0]
+["tcgen05_ld_16x256b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3360,null,null,3361,213],"837665bdd35882be4d92f2d0eae8fc4e9238e8769f1380a147705d183180db31",0]
+["tcgen05_ld_16x256b_x8_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3362,null,null,3363,213],"a8ff0f860a53f739dfd1e21363f7c91065c62e651e5a2965fd98f99043cff880",0]
+["tcgen05_ld_16x256b_x8_pure",0,0,"sm_100a","+ptx86","sm_100a",[3364,null,null,3365,213],"22cf3f09dc830a7191f92d7109d9d4f2fccc04b083dbf53ba185a78813756de4",0]
+["tcgen05_ld_16x256b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3366,null,null,3367,213],"878ab01e5cc551a57c4bbc2c0564c622dc0cfa33ddbc6d7d63e18e7dfb2a6b3e",0]
+["tcgen05_ld_16x32bx2_x128_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3368,null,null,3369,213],"dbabc74b2b56f849661005b5f03dda72fb6b1a0a8be19b86f4a6507d6675cdff",0]
+["tcgen05_ld_16x32bx2_x128_raw",0,0,"sm_100a","+ptx86","sm_100a",[3370,null,null,3371,213],"b63604c2adcea23e27aa6f3d91cfc5125473d3e23e2ef37a598b6eefc1f9c524",0]
+["tcgen05_ld_16x32bx2_x16_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3372,null,null,3373,213],"31278c5b529e16be3699255c25be669689cb04e925a6087b3f6744c2880e0965",0]
+["tcgen05_ld_16x32bx2_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3374,null,null,3375,213],"893f129d40134024286a328df4c1553e85f7a61c0d82fa612f87c29f3378b1cb",0]
+["tcgen05_ld_16x32bx2_x1_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3376,null,null,3377,213],"0a6019c84245cc90860dae382c9fa1b576bb4c1f2b18f2af67821c1f5e49be2e",0]
+["tcgen05_ld_16x32bx2_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3378,null,null,3379,213],"bc105e74acef425efdc55ef063a63a746615f473ef866bdf980b160d0e81b1ab",0]
+["tcgen05_ld_16x32bx2_x2_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3380,null,null,3381,213],"cfcf0a391898b032af501cc07e571712b38d7953c347aab3c5c67cc3a244e9b7",0]
+["tcgen05_ld_16x32bx2_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3382,null,null,3383,213],"e32629f5c1f2e9e73f9e62dc31b30c42023b066a207081345981be23994216ed",0]
+["tcgen05_ld_16x32bx2_x32_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3384,null,null,3385,213],"219f8060e97f0036c2560d1c243fa5691f9f1fe32356e20de9c3f617c98058c7",0]
+["tcgen05_ld_16x32bx2_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3386,null,null,3387,213],"109548b4c987432c9a2ea180505eac877442db2ca94c94f8d9a43b6375206033",0]
+["tcgen05_ld_16x32bx2_x4_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3388,null,null,3389,213],"7df4f0f8bd40fba2711f2280fa387f113905dbec2cdacedc4e108f9011e09175",0]
+["tcgen05_ld_16x32bx2_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3390,null,null,3391,213],"c620f53342e697c3e8bfcdddacf534fd2be16644fd9f072e4a9db67a1c56c693",0]
+["tcgen05_ld_16x32bx2_x64_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3392,null,null,3393,213],"9ef5c3cb09ee19ac96be42008b77179ae144df039c16be2c45b236424f2fee60",0]
+["tcgen05_ld_16x32bx2_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3394,null,null,3395,213],"fb7dce647f1b1a16435e4bec7bf99cc1db8da2025db6cdca65fd2ef63180b973",0]
+["tcgen05_ld_16x32bx2_x8_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3396,null,null,3397,213],"403e9d51b1dd77b0d380895556f0254e472a2bc70e921c88ab638df2f2871bbd",0]
+["tcgen05_ld_16x32bx2_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3398,null,null,3399,213],"9da69e353656bc6327fbac928e067f29332bdd9fa84ad4b4430acba727c38dd3",0]
+["tcgen05_ld_16x64b_x128_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3400,null,null,3401,213],"07011818e32da1302027899d5f22afcd95c72843eb8664f0369c98c6c2386665",0]
+["tcgen05_ld_16x64b_x128_raw",0,0,"sm_100a","+ptx86","sm_100a",[3402,null,null,3403,213],"f38e0599395b83868320e5cea33c1c193429423ef2467f0d8291a5840acba360",0]
+["tcgen05_ld_16x64b_x16_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3404,null,null,3405,213],"9bf83f6b9b8e8bb8154cdd0fe4f213f946571c76099b1465494afe2d3f4bb186",0]
+["tcgen05_ld_16x64b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3406,null,null,3407,213],"a8cf18567f9d67027a2059c7436aee1ac0d73053bb2ba94041788db499c2d154",0]
+["tcgen05_ld_16x64b_x1_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3408,null,null,3409,213],"2eae3714a914aac75ade68ce9edbee05182881e58ba9cd1cdbf0a001aed7f33a",0]
+["tcgen05_ld_16x64b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3410,null,null,3411,213],"37206c3e1fa73ba2182148c2cbf39c4b82333fd3c045ef41eedfe31183f1f7df",0]
+["tcgen05_ld_16x64b_x2_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3412,null,null,3413,213],"e480647024b3ed294f181d655f5f531cfa3beaa0b3412c8782083d28ff7bf23b",0]
+["tcgen05_ld_16x64b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3414,null,null,3415,213],"09200e4f5f97224c46000cf8d095018859d8e70c13c78d71d0cf48686ba9edd0",0]
+["tcgen05_ld_16x64b_x32_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3416,null,null,3417,213],"abff950062a9b2ea38e6014675f012916557dfe72ae9e92d13a851b88f398c33",0]
+["tcgen05_ld_16x64b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3418,null,null,3419,213],"6797ddd81bb4008683af8bed4cc3763a13ade861d8f077d463e96df25d0b9f43",0]
+["tcgen05_ld_16x64b_x4_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3420,null,null,3421,213],"6ff4456759059444383591483ddda7760a790c64e6b2a1327711734a206fb3d0",0]
+["tcgen05_ld_16x64b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3422,null,null,3423,213],"9e9b9567f49c65490bac9de958134db998b1da395f6bb187736971533ff1cfcc",0]
+["tcgen05_ld_16x64b_x64_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3424,null,null,3425,213],"ef02170ee165d8650c8376cf13681a2b9fe21bf239ab5ab2fb7cdffb624dada5",0]
+["tcgen05_ld_16x64b_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3426,null,null,3427,213],"cc6e67be709abb29038c665ab6638e2128c52273c29d4d9d38c7073e9257a92f",0]
+["tcgen05_ld_16x64b_x8_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3428,null,null,3429,213],"983206c8dac61a038a747be6884fc1d51f05676eb171cf8eac8df4c9a395eb8c",0]
+["tcgen05_ld_16x64b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3430,null,null,3431,213],"fbc560aa7135a32340124ac5920cd2f15d02e94791aa56a43f7a91372b7c1300",0]
+["tcgen05_ld_32x32b_x128_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3432,null,null,3433,213],"9931c967b5e6ef2630d2ad690ba47a03751a6afe699ef3c12942e29002987300",0]
+["tcgen05_ld_32x32b_x128_raw",0,0,"sm_100a","+ptx86","sm_100a",[3434,null,null,3435,213],"8200b6f5f8cb9871c8840f6897659bfb31166911c747e515348062bd26d9c6b0",0]
+["tcgen05_ld_32x32b_x16_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3436,null,null,3437,213],"216433f1d7c7150a04e237a5e80156c682a8e4da80d5a319b6122294fd038cd4",0]
+["tcgen05_ld_32x32b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3438,null,null,3439,213],"c51495cb19f74f582386fa08df3694744caf2ca5439cb5def556a3085f13ecb8",0]
+["tcgen05_ld_32x32b_x1_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3440,null,null,3441,213],"a9e391bbd0bce0b6080674f1f6e70823abdb80af7218041a8a5d5959ca7a055a",0]
+["tcgen05_ld_32x32b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3442,null,null,3443,213],"4b5e4aa32becf8d746c7b953a331627f45d2d0e95affc227135af5cfeb867ff6",0]
+["tcgen05_ld_32x32b_x2_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3444,null,null,3445,213],"4a72966641f344b80f945ed26f3d3c955924167f4c3708c135d727059cffd22e",0]
+["tcgen05_ld_32x32b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3446,null,null,3447,213],"2de97ea9929d5e70342aa3d833c174d24c38a8231a4e638eb39d4c711d71d84a",0]
+["tcgen05_ld_32x32b_x32_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3448,null,null,3449,213],"17f8c34f153a65b7c62061809f1d2c8140907912b3ddb6d1b4aeb36861c701a0",0]
+["tcgen05_ld_32x32b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3450,null,null,3451,213],"e2c461140bb0e21ac4fee9a3e73bdd2cfe85db4006d337c3b382c6b6d040e847",0]
+["tcgen05_ld_32x32b_x4_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3452,null,null,3453,213],"6f3b5242f59141a675b38a1cbb79fee8edbc35c6c6e20b0cbc3f695a8d9f2def",0]
+["tcgen05_ld_32x32b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3454,null,null,3455,213],"d99a8b40848ad6ea8b363a945d503ec721df6cf6cb577950dbb38c9fdebcc4ac",0]
+["tcgen05_ld_32x32b_x64_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3456,null,null,3457,213],"be7ba983544c235bf0774b4240dcb16fef8dd4885ef2b537532c34c6dc5bf43c",0]
+["tcgen05_ld_32x32b_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3458,null,null,3459,213],"00fb7bd01b4d8f591ab157d070987fbda9796f796677d2f37af524df75a50076",0]
+["tcgen05_ld_32x32b_x8_pack16",0,0,"sm_100a","+ptx86","sm_100a",[3460,null,null,3461,213],"989c9ee7dfc678435385ca04259af672f082ed90e57115fb702347896a835f0b",0]
+["tcgen05_ld_32x32b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3462,null,null,3463,213],"2f420dd0df448a90c699dc7cbc2bfbfbc2dac07801bcb1fda91a1825a1f2df5f",0]
+["tcgen05_load_wait",0,0,"sm_100a","+ptx86","sm_100a",[3464,null,null,3465,213],"b30c9c426b082c49c5e190eb53b737f397eddf25538cde6d0bff9b4cfdeb97a1",0]
+["tcgen05_mma_e2m1",0,0,"sm_100a","+ptx86","sm_100a",[3466,null,null,3467,213],"fe09e67c024b7af2feb72c71ecc0b627818a24f595bb82ecb9cd0cb373bcc435",0]
+["tcgen05_mma_e2m1",0,0,"sm_100f","+ptx88","sm_100f",[3468,null,null,3469,1076],"3c676a0bb27027ea49218242862c3ac29f276b35da5c099fd3040d62d034d801",0]
+["tcgen05_mma_e2m1",0,0,"sm_101a","+ptx86",null,[3470,null,null,3471,null],"441eb5aa3044ce83bca46b70f39581a865b9aded2d2b0f5ddb850c6e56efd888",0]
+["tcgen05_mma_e2m1",0,0,"sm_101f","+ptx88",null,[3472,null,null,3473,null],"eb0d3a1adb2544213817e60e5a6997355a16a091c2bde946debf6d92d7e651ec",0]
+["tcgen05_mma_e2m1",0,0,"sm_103a","+ptx88","sm_103a",[3474,null,null,3475,1079],"bdf81fc8df36c9a6bb62932d9cf674865261efa3bb2731e0849b86a14d2dfbd0",0]
+["tcgen05_mma_e2m1",0,0,"sm_103f","+ptx88","sm_103f",[3476,null,null,3477,1082],"7f31a8b4c1a4e3c7acc70f778a76abd0b98f9b42794046cc5576a6a4d285357f",0]
+["tcgen05_mma_e2m1",0,0,"sm_110a","+ptx90","sm_110a",[3478,null,null,3479,1085],"ce04d6d2c144f655e679cccd4f1a416e726ed52f07d10afda3625348c577b460",0]
+["tcgen05_mma_e2m1",0,0,"sm_110f","+ptx90","sm_110f",[3480,null,null,3481,1088],"385bf06a31fd2fac8c73808f976c6d0b2f0b1aa204630ec3fb6761c7fd7670fa",0]
+["tcgen05_mma_e2m3",0,0,"sm_100a","+ptx86","sm_100a",[3482,null,null,3483,213],"9c828951addb4d3c383bb0b29a6eb9fb44834c61d729f459ef868dc5e3ebc831",0]
+["tcgen05_mma_e2m3",0,0,"sm_100f","+ptx88","sm_100f",[3484,null,null,3485,1076],"5ed4e15f34ad3e7464681be1f72883a73d2ab8f3cd7a649f8b9728c7f9f35cb8",0]
+["tcgen05_mma_e2m3",0,0,"sm_101a","+ptx86",null,[3486,null,null,3487,null],"fc67a2855bc29424c69f657a9b876d171d81f5346ad3c664df56b8524de0fa28",0]
+["tcgen05_mma_e2m3",0,0,"sm_101f","+ptx88",null,[3488,null,null,3489,null],"a331553b8f5eded4af050fb911fd30c07b8ce8d587997bd5ebc69a155935ce8c",0]
+["tcgen05_mma_e2m3",0,0,"sm_103a","+ptx88","sm_103a",[3490,null,null,3491,1079],"bd0a3e2bcc8545a476821d600d359ab404503e6dd238486b751214aa1eb90b03",0]
+["tcgen05_mma_e2m3",0,0,"sm_103f","+ptx88","sm_103f",[3492,null,null,3493,1082],"44bd03264ad795d51820e7c26deedf2d88b4b35359bbccc5cfbbd7b11ecfd9a7",0]
+["tcgen05_mma_e2m3",0,0,"sm_110a","+ptx90","sm_110a",[3494,null,null,3495,1085],"4d0bff0022923b7dad7fe1bc87004eaea9e57d96c31c931b7c54c033c475c6a8",0]
+["tcgen05_mma_e2m3",0,0,"sm_110f","+ptx90","sm_110f",[3496,null,null,3497,1088],"1be1a2dde853fa6c5c9a8cc0f22f6a65ed7ba8fd454ae3306a7aed28bc215d38",0]
+["tcgen05_mma_e3m2",0,0,"sm_100a","+ptx86","sm_100a",[3498,null,null,3499,213],"912ac08300c253cd406696b7125cdb3ff00cd9d4cc1d825c41ec2a3960768544",0]
+["tcgen05_mma_e3m2",0,0,"sm_100f","+ptx88","sm_100f",[3500,null,null,3501,1076],"f021bcea3874c7edfd0504092e758fb8a52afae2471414602bc38df06f55232e",0]
+["tcgen05_mma_e3m2",0,0,"sm_101a","+ptx86",null,[3502,null,null,3503,null],"1c1fc28cf55726086884323697e3219277f64019062068391d93496b6f850cf6",0]
+["tcgen05_mma_e3m2",0,0,"sm_101f","+ptx88",null,[3504,null,null,3505,null],"fdb2ffc918e2455e9e3fadcfd21b69774e4f01f31af72d90c601b2c36685bee2",0]
+["tcgen05_mma_e3m2",0,0,"sm_103a","+ptx88","sm_103a",[3506,null,null,3507,1079],"286d10fa824403c742f2f99556347a1433598a3c8ba4f026b3d04621db57a70a",0]
+["tcgen05_mma_e3m2",0,0,"sm_103f","+ptx88","sm_103f",[3508,null,null,3509,1082],"2044c9b3af54ad638c91b54d2b8c5a93f864d600521a5dce6454ba3f88d74f33",0]
+["tcgen05_mma_e3m2",0,0,"sm_110a","+ptx90","sm_110a",[3510,null,null,3511,1085],"c4e9113e773e8bb664048aa514477b81488f4263829d2e63cff0265deb920ed5",0]
+["tcgen05_mma_e3m2",0,0,"sm_110f","+ptx90","sm_110f",[3512,null,null,3513,1088],"b81b430aee639d3d47e0606cf328d01402f38e3caf0d3b5d63c42554e5c4cd18",0]
+["tcgen05_mma_e4m3",0,0,"sm_100a","+ptx86","sm_100a",[3514,null,null,3515,213],"39b8cc63992c7914fa390fcd01d5a3c4e3386cc3d69c1248c20c32cd064187a5",0]
+["tcgen05_mma_e4m3",0,0,"sm_100f","+ptx88","sm_100f",[3516,null,null,3517,1076],"4108a1c28bb1549925cf3c654dd1b4936ffd77308d3930464c675750be9e08ce",0]
+["tcgen05_mma_e4m3",0,0,"sm_101a","+ptx86",null,[3518,null,null,3519,null],"67eae2009f8fae2f42224e0f5074d6e25c59d472b5143d587be942b06f1fd19b",0]
+["tcgen05_mma_e4m3",0,0,"sm_101f","+ptx88",null,[3520,null,null,3521,null],"783e01493fedee5f84b4e29c5e61b7104091f9a44d9cd65c23d3c4b44265b9c6",0]
+["tcgen05_mma_e4m3",0,0,"sm_103a","+ptx88","sm_103a",[3522,null,null,3523,1079],"80db4d6057dca68c4cec1e80a1359b8401400d148659e5c9b0b9720b9557ccba",0]
+["tcgen05_mma_e4m3",0,0,"sm_103f","+ptx88","sm_103f",[3524,null,null,3525,1082],"afe318203de7ab011158396d8de134b84715d8cbe49c7b2026576a1d396ae855",0]
+["tcgen05_mma_e4m3",0,0,"sm_110a","+ptx90","sm_110a",[3526,null,null,3527,1085],"b0a3d6ed1164a555babbc5f639791d84872fca5a89c36b144bd781e5b3fcdd2b",0]
+["tcgen05_mma_e4m3",0,0,"sm_110f","+ptx90","sm_110f",[3528,null,null,3529,1088],"733015a2fcb748321ff8411abb3211a7bfec9646ff7429d50d006bd40eb5edca",0]
+["tcgen05_mma_e5m2",0,0,"sm_100a","+ptx86","sm_100a",[3530,null,null,3531,213],"4f05cdc091d425f3320e341eafe157590e60051e7f2468688c3bcb74e428c14a",0]
+["tcgen05_mma_e5m2",0,0,"sm_100f","+ptx88","sm_100f",[3532,null,null,3533,1076],"c603f5b976c9b52daf3c70ea3f557f87e9f33104f784b2d198dff8ed63bed577",0]
+["tcgen05_mma_e5m2",0,0,"sm_101a","+ptx86",null,[3534,null,null,3535,null],"8e7ed65db00588df273706c68cf00260aab42456e35142478d182d392c8f5f8f",0]
+["tcgen05_mma_e5m2",0,0,"sm_101f","+ptx88",null,[3536,null,null,3537,null],"6e88cf369440780047b7b9f9cf7e7c8916c1b1407441e88ba9e77922d1d14b94",0]
+["tcgen05_mma_e5m2",0,0,"sm_103a","+ptx88","sm_103a",[3538,null,null,3539,1079],"c8d8bb7f189f340959c117d9b9b6b3a4034db47be8ce1ac25129692a476170b8",0]
+["tcgen05_mma_e5m2",0,0,"sm_103f","+ptx88","sm_103f",[3540,null,null,3541,1082],"d7b28fa8495a7ec234e16ac29d5461fedc96e92055a96f2f6ff497ca7432d8e2",0]
+["tcgen05_mma_e5m2",0,0,"sm_110a","+ptx90","sm_110a",[3542,null,null,3543,1085],"0eca17a4a65ae88847891ac3fcb709452145c69ffbed8f7f2a8a203ac017b91b",0]
+["tcgen05_mma_e5m2",0,0,"sm_110f","+ptx90","sm_110f",[3544,null,null,3545,1088],"0340534ce970044bb729358b5ae97a947c62e315a66e463568b06c7af450f0c2",0]
+["tcgen05_mma_f16",0,0,"sm_100a","+ptx86","sm_100a",[3546,null,null,3547,213],"ca642d2f51eab21201ebc3aa9330e2951a0dc3e215346d9b764e862899a98040",0]
+["tcgen05_mma_f16_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3548,null,null,3549,213],"a16b9e429cac7addd93ee6a354e1664efce60d764962fda41798a96ff1c2cd97",0]
+["tcgen05_mma_shared",0,0,"sm_100a","+ptx86","sm_100a",[3550,null,null,3551,213],"e07ecf043868c4eef4c3fa85bd8f16cca62eb6c7f3f273b4738292c92be57c6d",0]
+["tcgen05_mma_shared",0,0,"sm_100f","+ptx88","sm_100f",[3552,null,null,3553,1076],"c90a2de58e265262b5ae3a3706447d28aa6568bcd97e07aa569c4ef45f0d52b2",0]
+["tcgen05_mma_shared",0,0,"sm_101a","+ptx86",null,[3554,null,null,3555,null],"e8a0bbd004b2ae2905c17cb9dc2138158edb8148eee979a217000bec4a5bc052",0]
+["tcgen05_mma_shared",0,0,"sm_101f","+ptx88",null,[3556,null,null,3557,null],"95194db70ce9b75b8711a4c2f4159a2eb620bbf8282a5db464dc653fbdf3b905",0]
+["tcgen05_mma_shared",0,0,"sm_103a","+ptx88","sm_103a",[3558,null,null,3559,1079],"49297a41f31019c496d78722bb7c9577b687718b5c765413c02b352987ce4509",0]
+["tcgen05_mma_shared",0,0,"sm_103f","+ptx88","sm_103f",[3560,null,null,3561,1082],"8bbe8cc31b36f40c8704b4fc61c70aa439860da2fcffc3320f3de6ca3ffc1f65",0]
+["tcgen05_mma_shared",0,0,"sm_110a","+ptx90","sm_110a",[3562,null,null,3563,1085],"dbea57143dbe256cc6595b114e1e7a2480fc6c3335ff4ee5988babb611b051fe",0]
+["tcgen05_mma_shared",0,0,"sm_110f","+ptx90","sm_110f",[3564,null,null,3565,1088],"71b2952688e95182ddf908f73037fd549d43cc4c82a1148f108d57ca18197b2c",0]
+["tcgen05_mma_sp_shared",0,0,"sm_100a","+ptx86","sm_100a",[3566,null,null,3567,213],"b7e5022d3cd2c912e340c907b3c3217fc859bfc4366c941e99d6c2c266978936",0]
+["tcgen05_mma_sp_shared",0,0,"sm_100f","+ptx88","sm_100f",[3568,null,null,3569,1076],"239a88ffb76b55121af020e18aa8aa4c398183e88b9f8a0fd84a3457d9d06ba7",0]
+["tcgen05_mma_sp_shared",0,0,"sm_101a","+ptx86",null,[3570,null,null,3571,null],"4c4df17cc711d4ffe32447bddf8b6f0fa0b2af04e4cb3e15b1371b1644660c90",0]
+["tcgen05_mma_sp_shared",0,0,"sm_101f","+ptx88",null,[3572,null,null,3573,null],"81e3bc921df574c74eba4df1bb13fd2cbdd088c335495487091547caaf4ac61f",0]
+["tcgen05_mma_sp_shared",0,0,"sm_103a","+ptx88","sm_103a",[3574,null,null,3575,1079],"daf75e48ad6e4a692200ec7914503ca143b63641e642bfe084f21ae4c7044024",0]
+["tcgen05_mma_sp_shared",0,0,"sm_103f","+ptx88","sm_103f",[3576,null,null,3577,1082],"34c508f8af9ff03240ba8c301524a7f0fa4333d6dda280b5f0985a99aaec876b",0]
+["tcgen05_mma_sp_shared",0,0,"sm_110a","+ptx90","sm_110a",[3578,null,null,3579,1085],"3d18091e3b05f64118e9a8085de9d40525ec8ab284cde2f22c0985de29297d75",0]
+["tcgen05_mma_sp_shared",0,0,"sm_110f","+ptx90","sm_110f",[3580,null,null,3581,1088],"1aef7484d826a5f5e939be99a32a70769ae6776301e80a9dabee9f30d82881b1",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_100a","+ptx86","sm_100a",[3582,null,null,3583,213],"2f3a1bc7640c2c00597b7f0b7e5b8fe2d4ab3aba9bc36b74d19a8f7190e03974",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_100f","+ptx88","sm_100f",[3584,null,null,3585,1076],"459471846ae4593a0a42e6aa68fb4743d5907a73977cc8ff2234cfaa99075c43",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_101a","+ptx86",null,[3586,null,null,3587,null],"5dfc80376cf7ac25c5ed95411110c32b7e4c40eedc74d7ba8499216468256340",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_101f","+ptx88",null,[3588,null,null,3589,null],"8ffe3fd1c8b700b8eb69f79e8ea734ecdca9f72f120669ff5380e48aeb6e204c",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_103a","+ptx88","sm_103a",[3590,null,null,3591,1079],"38b7e7375576b6e02315739b9799dfaf8ecb2f329bdc444a3405718154478149",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_103f","+ptx88","sm_103f",[3592,null,null,3593,1082],"10ef0246a01e7c35883d2a3ead978e4dbf5d7a8b29378021a28ef27407a2efcf",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_110a","+ptx90","sm_110a",[3594,null,null,3595,1085],"6fb664a9aaa5d555acae20e916b526a236426e01df0122d9685e9a4d00ad561e",0]
+["tcgen05_mma_sp_tensor",0,0,"sm_110f","+ptx90","sm_110f",[3596,null,null,3597,1088],"3858c1ca375c58ff52af1164c892506066e095dbe38176f779fa3062b6bf0b43",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_100a","+ptx86","sm_100a",[3598,null,null,3599,213],"102eb3d3a69e34a29ebca32c3394d28a9419181509e4560e425cf66e7c3eb64c",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_100f","+ptx88","sm_100f",[3600,null,null,3601,1076],"2034fab4d4487fb4269fb778ef443614c92dbd97ca1ffad8ef69c1da48a51c8a",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_101a","+ptx86",null,[3602,null,null,3603,null],"b2a2becc1307f4d64782195b20aa520fdda07fbf7f6737215ad8e873b4dfb1c9",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_101f","+ptx88",null,[3604,null,null,3605,null],"35748093a4634fcc79b15268f8d0792515c68981d32c929b8590a4879e7102bb",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_103a","+ptx88","sm_103a",[3606,null,null,3607,1079],"09af7b1ec193fcafad70e3c233a08591a8a0e05215bf4e27e34dc489b86613a7",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_103f","+ptx88","sm_103f",[3608,null,null,3609,1082],"85f55b096b800454914b87e9e0ffdce3b0e814fdde95d68b18060e0d73b639fa",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_110a","+ptx90","sm_110a",[3610,null,null,3611,1085],"5d13811ebf353e1cb9a72839600cbbfe2b600ef4efa9ec02359665ababfb756d",0]
+["tcgen05_mma_sp_tensor_ashift",0,0,"sm_110f","+ptx90","sm_110f",[3612,null,null,3613,1088],"763f52a2faa74bf010875cee1e5e7a5236f8f0a6152d6e9ee09c02ef9a20c27f",0]
+["tcgen05_mma_tensor",0,0,"sm_100a","+ptx86","sm_100a",[3614,null,null,3615,213],"d5d1f623e9c323e9d1a4d632b69f1b62c5a17ea17ed2265c9bce359022f06ee0",0]
+["tcgen05_mma_tensor",0,0,"sm_100f","+ptx88","sm_100f",[3616,null,null,3617,1076],"1c9958cdd3e1f73d892772bf47ea23706b2265999f17fed0e4967d5e89ea89c3",0]
+["tcgen05_mma_tensor",0,0,"sm_101a","+ptx86",null,[3618,null,null,3619,null],"f05b76be2f4402593fffdf25042f201ce39b4441e9155511bb068736773a60f9",0]
+["tcgen05_mma_tensor",0,0,"sm_101f","+ptx88",null,[3620,null,null,3621,null],"41000cc9500252134f27248489895adb8f4283281e742034306941645a5d5a4d",0]
+["tcgen05_mma_tensor",0,0,"sm_103a","+ptx88","sm_103a",[3622,null,null,3623,1079],"d3c5e99ed300d5848ad53d69fa6f68052e94e912b286385367886bc5372a37f6",0]
+["tcgen05_mma_tensor",0,0,"sm_103f","+ptx88","sm_103f",[3624,null,null,3625,1082],"6f77377f64ebe43433fd437b0db0b2c02c863a1cddf1dadc1ae845b19d33c9ec",0]
+["tcgen05_mma_tensor",0,0,"sm_110a","+ptx90","sm_110a",[3626,null,null,3627,1085],"e74b4d2ab0b409335e6cf2cea93fe300613ab6cc6bbee82be58d60094452f5be",0]
+["tcgen05_mma_tensor",0,0,"sm_110f","+ptx90","sm_110f",[3628,null,null,3629,1088],"0b9cb19d0c8a6a151f4fffefe9c3482e558f96be87e7d716432b3da266c344a4",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_100a","+ptx86","sm_100a",[3630,null,null,3631,213],"7af2546aeaf5ce85dc104667c6c14cf35ab20fcbe47d79af8decb29a3481f82f",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_100f","+ptx88","sm_100f",[3632,null,null,3633,1076],"a5ed301f6a84550ecf84c023f93336d2ee9cdf0a1461d7a58df0cd93e4c77ba4",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_101a","+ptx86",null,[3634,null,null,3635,null],"79cdda209dd936ee1682fe6282fa0766464aa43e6b9f55ed1ac49106550583c9",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_101f","+ptx88",null,[3636,null,null,3637,null],"4f5a213ba949eb4ced3c408e6c1b8bc29888e4443c5e52781cdbf8fc1b704a43",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_103a","+ptx88","sm_103a",[3638,null,null,3639,1079],"1fd970b8efffe1b9506109caa756e45f84bfaedb58cab4406d75b260c199f07a",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_103f","+ptx88","sm_103f",[3640,null,null,3641,1082],"1047f10d233f0dad1668a24c3a0b9948ea30d1504af204ac38d89fc61dbd3f38",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_110a","+ptx90","sm_110a",[3642,null,null,3643,1085],"4e933c3f3d11531bfe7d12ca1b7bd2308fe4ec4f6970b3dd0d233cb226483375",0]
+["tcgen05_mma_tensor_ashift",0,0,"sm_110f","+ptx90","sm_110f",[3644,null,null,3645,1088],"9125e462d764c08a55410574a1ec6c5baf05003ebf36576f22175d25224dbc80",0]
+["tcgen05_mma_ws_bf16",0,0,"sm_100a","+ptx86","sm_100a",[3646,null,null,3647,213],"5a10a3162a726722bd7e7db192f2f0c791fd9871a469cdb24130e58179799a5f",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_100a","+ptx86","sm_100a",[3648,null,null,3649,213],"4abf152a4762e72a53986df763f062c18a5b661c7b1be816008ba598d02b8ae4",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_100f","+ptx88","sm_100f",[3650,null,null,3651,1076],"e6d0755d3f7238bd0ca112770d6933dd6f618fff3e966dbfbdd0c9becc3aa99a",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_101a","+ptx86",null,[3652,null,null,3653,null],"feeef1a7ca31a44719a0e1f36b006fed50f31b6d91c357b4400775ee1846496c",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_101f","+ptx88",null,[3654,null,null,3655,null],"cc7560d644a84142516e6e9004f3845b573a67b032f134759f1ca97c672bc078",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_103a","+ptx88","sm_103a",[3656,null,null,3657,1079],"8af2aac98481827db0cc2e0a283dd084f929587eef1feeec2c91ec26e32bc706",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_103f","+ptx88","sm_103f",[3658,null,null,3659,1082],"bc9bc735a01406102fb8546e270421ba208d41872c2f3fb536f12ed46423b761",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_110a","+ptx90","sm_110a",[3660,null,null,3661,1085],"c3e6bd9ffdb44a15421a7d129e7f96a3865a0be9238081672b4b3294cf8efd84",0]
+["tcgen05_mma_ws_e2m1",0,0,"sm_110f","+ptx90","sm_110f",[3662,null,null,3663,1088],"97ed743d7af1080cbcd87e4866f03fba38f3be44236266aa3bcadeb49a76b623",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_100a","+ptx86","sm_100a",[3664,null,null,3665,213],"3abcc12e8a50c35d467c9c19251bcf9ac8b1552b472ff46eb09714c177c4e6b5",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_100f","+ptx88","sm_100f",[3666,null,null,3667,1076],"c04bd9518d7379cfc89db7f5bef97927903f3a61445ac48a06c0cb3ab2ff2cf8",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_101a","+ptx86",null,[3668,null,null,3669,null],"b493189d90fdf93a4bb0be5389d1760314dea2b519bd24c7a120f14671de5559",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_101f","+ptx88",null,[3670,null,null,3671,null],"48fdc0f1140ba328cc4deab65e84a84a2498f7495e0b0af60a99c312be141a02",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_103a","+ptx88","sm_103a",[3672,null,null,3673,1079],"10093beef8f22db9abb8a497f4d16aebe028d0ca59ba40936d007651477333b0",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_103f","+ptx88","sm_103f",[3674,null,null,3675,1082],"02091fea4036fc3c413acd3e0ff81eef6067097086a5da08839ec5a6ead9046a",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_110a","+ptx90","sm_110a",[3676,null,null,3677,1085],"213b7caf44494dd4b4e2b52f4018b4968f9bd4ddd2c0022ac7309ddca1a6f2a5",0]
+["tcgen05_mma_ws_e2m3",0,0,"sm_110f","+ptx90","sm_110f",[3678,null,null,3679,1088],"9b32848b820778404e68ad7f12f4808be5f4530d09dd3136a03ac29ad0f084f0",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_100a","+ptx86","sm_100a",[3680,null,null,3681,213],"0e4798aae840bc7878a9b5e57dcd1b55c6d78540c9227be6f98c49d43d7eaf0d",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_100f","+ptx88","sm_100f",[3682,null,null,3683,1076],"cf0b0b15824c1200338b418e621bc5726cbdc12f540d281a320af7ec17b52dd8",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_101a","+ptx86",null,[3684,null,null,3685,null],"be94dfe40c8eff4ab0deb0dd27ec57f3b93723d26c7064f419013091fb81a9d2",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_101f","+ptx88",null,[3686,null,null,3687,null],"32534b206e731c4909ab327fca6fcf227fff7ed83556cc450cc54be259c75962",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_103a","+ptx88","sm_103a",[3688,null,null,3689,1079],"c84132b7c230a38fada2facc0e7d402b0e8ad91bdd64bc83fefd2c240638e554",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_103f","+ptx88","sm_103f",[3690,null,null,3691,1082],"c8e053d3773a59367cc57f0c78827ee9754de2d011bb1c480ff9725ca674cce6",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_110a","+ptx90","sm_110a",[3692,null,null,3693,1085],"8f77592728ca18d4c42574b021ad478a0bb739a26be2b5c6d4f936eefeac5c3a",0]
+["tcgen05_mma_ws_e3m2",0,0,"sm_110f","+ptx90","sm_110f",[3694,null,null,3695,1088],"f05e5cc372e410d636a00435dacfcd2bffc31944bd36977221c794930aee1d15",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_100a","+ptx86","sm_100a",[3696,null,null,3697,213],"9960abca4c7f5c08d79f07f30151d5127b806608d4b442c66ad36a3865733fae",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_100f","+ptx88","sm_100f",[3698,null,null,3699,1076],"74c6980109d53b9066f63740ad3bcae77ae74eb64ae407f5e90749bf39485370",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_101a","+ptx86",null,[3700,null,null,3701,null],"8d1ca918160c4ac98490126c40d0ec9611df229ab62a139ae7dded2fa690833b",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_101f","+ptx88",null,[3702,null,null,3703,null],"153eb487d4e776679a1da38237d61bcafa0dbcfb3af10c978e32637e86394b68",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_103a","+ptx88","sm_103a",[3704,null,null,3705,1079],"7393a7cad319cfad86beb55accd6df4e532f7347ee292289f6fc1c0c332219ac",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_103f","+ptx88","sm_103f",[3706,null,null,3707,1082],"2e438effd996d36db970121870ddd7fb977baa0db4bfb0d087556182dd24dece",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_110a","+ptx90","sm_110a",[3708,null,null,3709,1085],"c41565e0e582a18f69ab01b6d85b7edd9cb2e7ba3f6fb98807daa94159bbb50a",0]
+["tcgen05_mma_ws_e4m3",0,0,"sm_110f","+ptx90","sm_110f",[3710,null,null,3711,1088],"94af5d5cff98489ecf8fc197115a73ee434fb40f6b568fffd5fdaa6d0e7bcb24",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_100a","+ptx86","sm_100a",[3712,null,null,3713,213],"f03ef9620d9bdf46b93b8dba488309e1d3f63dd8c6ee7097a47d2410221727d2",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_100f","+ptx88","sm_100f",[3714,null,null,3715,1076],"8b798a41a65beb5d823fda4788751376cfd7808fa420c01f91a0d5e2a184a68e",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_101a","+ptx86",null,[3716,null,null,3717,null],"75a3a72f702b9e57ba068ff47683f693a0e57a21aa883e35e78fe93cec6d25e5",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_101f","+ptx88",null,[3718,null,null,3719,null],"d28472f2892dd01bff875f51ca09b7d0bc0e25445c9f8f0b8832403733ff9ae8",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_103a","+ptx88","sm_103a",[3720,null,null,3721,1079],"f3ee7b2b1f9896f6fa6f54e0598929aebe6bf6064854e93906605087074e61c8",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_103f","+ptx88","sm_103f",[3722,null,null,3723,1082],"c65c5235c700abadb39111d4ddf4a899118e345e55ecd972dfa0d7971f46861e",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_110a","+ptx90","sm_110a",[3724,null,null,3725,1085],"0b3006e2e0cb697fb00fbb1ac56ad0689b77150e13059a2a1dabea53a8bb4b85",0]
+["tcgen05_mma_ws_e5m2",0,0,"sm_110f","+ptx90","sm_110f",[3726,null,null,3727,1088],"a12d97887750a9882b2923fcd8c539ebb38cf29e8882e0f8f5905aa7abbbb0ba",0]
+["tcgen05_mma_ws_f16",0,0,"sm_100a","+ptx86","sm_100a",[3728,null,null,3729,213],"eb92b68c7e1508751bc7b10bc40bd40ed0b9eae6db8c2deaa07d2d6c48994fdb",0]
+["tcgen05_mma_ws_shared",0,0,"sm_100a","+ptx86","sm_100a",[3730,null,null,3731,213],"c675f5cf38aa47a5ca02fa4f59d4328fa1832eabae55b6f1eb3a40da7ff89e99",0]
+["tcgen05_mma_ws_shared",0,0,"sm_100f","+ptx88","sm_100f",[3732,null,null,3733,1076],"15d46b9456b92c6b74e810605ee039b6802aa7fdad9f45c6e32b952ad520357b",0]
+["tcgen05_mma_ws_shared",0,0,"sm_101a","+ptx86",null,[3734,null,null,3735,null],"b7d38b604c0fc197e59ee58dee9a455cefd50343761e2e93e14b13478fb5a476",0]
+["tcgen05_mma_ws_shared",0,0,"sm_101f","+ptx88",null,[3736,null,null,3737,null],"b339fa16e2e7abf51d6b7e15f0843a9c883d0be1c2104e20c0cbe8ce71bed3ae",0]
+["tcgen05_mma_ws_shared",0,0,"sm_103a","+ptx88","sm_103a",[3738,null,null,3739,1079],"566c766321ebf27719f94989f108cf25f05df60beb76fd1d60e56381240a20b2",0]
+["tcgen05_mma_ws_shared",0,0,"sm_103f","+ptx88","sm_103f",[3740,null,null,3741,1082],"5a2fe4ff270c954289d371f43ca1e3ce4a17996ebf1d396a021b6d09456f0171",0]
+["tcgen05_mma_ws_shared",0,0,"sm_110a","+ptx90","sm_110a",[3742,null,null,3743,1085],"c805446c8a56c6b0fb392236b61b23a18ca0851b10a0cf2c879076b6cb7fdb38",0]
+["tcgen05_mma_ws_shared",0,0,"sm_110f","+ptx90","sm_110f",[3744,null,null,3745,1088],"62c896883483d2068b3ba895f499bb54f916d6866998bdf4c46e9e6c758b18ba",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_100a","+ptx86","sm_100a",[3746,null,null,3747,213],"359027ef06737064662d0ae73794a475da3ddd8d796b0a76e13214aca872f074",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_100f","+ptx88","sm_100f",[3748,null,null,3749,1076],"f08921c5cf74115549301fe88411be91f05db3974b54da9f7073ee20831b73b9",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_101a","+ptx86",null,[3750,null,null,3751,null],"6f9633797cf971a8db08272e7eb48ea8e4185c1cab3c675d29e0c5dcf6be5a11",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_101f","+ptx88",null,[3752,null,null,3753,null],"4e82aad77ac1bbb51f25c65c8b70f87437eb9d01c412ca294f6d964e3f5cfd8d",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_103a","+ptx88","sm_103a",[3754,null,null,3755,1079],"eac5a254e97dfcb8431071bb76fd4d9cc977aa36283cae4644921bcda38ed9f0",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_103f","+ptx88","sm_103f",[3756,null,null,3757,1082],"fcb76f21d47e0dbbb765b3f9d397ba5357cdd51af5d766bcb60b9794671a5490",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_110a","+ptx90","sm_110a",[3758,null,null,3759,1085],"5a19eb9e0e17e07bb23174bc89a7f8ff944a63c13ac37d64f1edf70273e3f6fc",0]
+["tcgen05_mma_ws_shared_zero_col_mask",0,0,"sm_110f","+ptx90","sm_110f",[3760,null,null,3761,1088],"f39758380e1000d460f65ac1b74db4e0831fcdbbf98906087088ebccfc7dfa91",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_100a","+ptx86","sm_100a",[3762,null,null,3763,213],"f51bb21d70b0902302aa2d3f771e800e41f28820d05faa28abaf1c640667d739",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_100f","+ptx88","sm_100f",[3764,null,null,3765,1076],"4b41a9cfeb5d4afe384e441be7f49afdd37540bc464014edd20db3990a907c54",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_101a","+ptx86",null,[3766,null,null,3767,null],"3b4d38bcac486cc3ed660ae6befc1a3f4922214a9f4526685a13ed97bd0b1400",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_101f","+ptx88",null,[3768,null,null,3769,null],"e5fb999c7eaaffceec1813f726684fd5a45c9a3dcab06f940beae852e4f41664",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_103a","+ptx88","sm_103a",[3770,null,null,3771,1079],"839c6da900fc073b07126a16a9fe0a47b5a7ad92095e471ae5981d71ceaa4d39",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_103f","+ptx88","sm_103f",[3772,null,null,3773,1082],"16b364747a61649adf65bd6712905fddaa085f07d4688c6af0f4591fd0752438",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_110a","+ptx90","sm_110a",[3774,null,null,3775,1085],"f8c7b2888e651064af5ed18a9249ce26f5947204ca42b67487a96a4a9a47adcd",0]
+["tcgen05_mma_ws_sp_shared",0,0,"sm_110f","+ptx90","sm_110f",[3776,null,null,3777,1088],"db0a06273a05f3cd57d9d5a396be5c848fc44547abd949b804a3f09fb64bfdad",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_100a","+ptx86","sm_100a",[3778,null,null,3779,213],"9f2b341a46a46c4673860e978553077f57f1177e1ff9de607e8ebe63d6cba7f7",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_100f","+ptx88","sm_100f",[3780,null,null,3781,1076],"56ab6b0456e9e5271b8ff6d5c435a9a93cfcd60429edd764905f6e05c957245e",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_101a","+ptx86",null,[3782,null,null,3783,null],"e77231357508f2f6fb49b6cba99b9fb609d9a685932a89fe6ea63a98ad200fdb",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_101f","+ptx88",null,[3784,null,null,3785,null],"b591fc8ae5bee5d9e94f1b7f028cad264f4bb6cb8f76de2b40684b5b9ddaba3d",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_103a","+ptx88","sm_103a",[3786,null,null,3787,1079],"8bfdde0a75b8bb73c4b7d34569f4dce7e812c168a341fe09e0e1fe3ab1ad098d",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_103f","+ptx88","sm_103f",[3788,null,null,3789,1082],"430a53aac169f9841077394ecef05a0dfcf1f0e1dc2d85b41ac30f42f356e289",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_110a","+ptx90","sm_110a",[3790,null,null,3791,1085],"c48fce2588fef97dedbea0df8c252fa0a3a0b1a5abdad79fea38ea8ac36f0bc3",0]
+["tcgen05_mma_ws_sp_shared_zero_col_mask",0,0,"sm_110f","+ptx90","sm_110f",[3792,null,null,3793,1088],"90b48518a5c1fcc7bb90062d0573e7716226cc93a56a6418f2894585488d8412",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_100a","+ptx86","sm_100a",[3794,null,null,3795,213],"758eaac5df41a8047bb424eb385eb74f91d71e7737ec113b98822ab62b713dc3",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_100f","+ptx88","sm_100f",[3796,null,null,3797,1076],"ae455855072531d0f4731fbd8c70d7e10821d800d1e5e423768eccf6beca797b",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_101a","+ptx86",null,[3798,null,null,3799,null],"dcc8d20b3f055fbb1e3515543a788c221099f28cb8be27164648c5ebe13d508e",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_101f","+ptx88",null,[3800,null,null,3801,null],"cbb2727d89bccc6f97505db636d91628df353ca8721426f44a77f1bcac585801",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_103a","+ptx88","sm_103a",[3802,null,null,3803,1079],"2ad3545f378307875229a9bd795ce22605dd381f9c332a38ef803c195cd5a573",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_103f","+ptx88","sm_103f",[3804,null,null,3805,1082],"8ddcd2e5e9835641a5689383a3ea2b43ab6fc6b2eaf182c26ff0d8970a3a54d7",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_110a","+ptx90","sm_110a",[3806,null,null,3807,1085],"87c5a3b60ea452dc44ac5472476074c25ff581c115e8be2d5407e6a7dd533d17",0]
+["tcgen05_mma_ws_sp_tensor",0,0,"sm_110f","+ptx90","sm_110f",[3808,null,null,3809,1088],"4018b2cc25d799f3a3d86d739e3de2c9e8eb129c7a0e92f4f8eee9828772523b",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_100a","+ptx86","sm_100a",[3810,null,null,3811,213],"7a15eb572d9c2de5256f2ba8b676c3564e880ba6e2cae85f4cc77e471bef95a0",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_100f","+ptx88","sm_100f",[3812,null,null,3813,1076],"b35e154e2b0ced626ed793045d48877ba229fc098b7b03914cb975fd926e30d9",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_101a","+ptx86",null,[3814,null,null,3815,null],"6787e003ca405f4a5c1e1e1f72e8161fd409c8c5ec5cddc8b5453e622b8a3a0a",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_101f","+ptx88",null,[3816,null,null,3817,null],"7f4caf368b0330562f883aeffd4a4493488b4eed152bba2bbd4dbde7512cb26a",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_103a","+ptx88","sm_103a",[3818,null,null,3819,1079],"1fc56bcbed22813305c5accf9a551dc7eb63aeb37ee954ab6ec039e961dba384",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_103f","+ptx88","sm_103f",[3820,null,null,3821,1082],"73a3fbffbe69d76be69019a7faa3e5465c51b63ed978d7ca53b0ea8a6db9aa6f",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_110a","+ptx90","sm_110a",[3822,null,null,3823,1085],"dbd2f50eb153ffceb66ced437409a4c5e15acf4204fe2f147e20831e3be014e1",0]
+["tcgen05_mma_ws_sp_tensor_zero_col_mask",0,0,"sm_110f","+ptx90","sm_110f",[3824,null,null,3825,1088],"c05b088c545475efd871513bd2f032c524af3fe8729332eae1aedb9f06499c77",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_100a","+ptx86","sm_100a",[3826,null,null,3827,213],"4716444c02adbbb751c30dc09f1002ff3510ab8461b55e683dde17a12484e784",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_100f","+ptx88","sm_100f",[3828,null,null,3829,1076],"497178591c443c2608a85d82ab83d9a047a7bea5fe578a5fe2705760f184d598",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_101a","+ptx86",null,[3830,null,null,3831,null],"86f6638db9671d9564016f006003fcafce2a10093e05db9a67fb8a553e1689f2",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_101f","+ptx88",null,[3832,null,null,3833,null],"bcde01244e40731724c705895822564c85f1cb7af7d68c99924423ddf4a15429",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_103a","+ptx88","sm_103a",[3834,null,null,3835,1079],"84bcfff74b9492f120d19d1db723363a282de5405d0aba5d4108af9a68616f37",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_103f","+ptx88","sm_103f",[3836,null,null,3837,1082],"64b395110a37bb3cce0fa6068401b30bc6b8be18312201335a6b99132850de0d",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_110a","+ptx90","sm_110a",[3838,null,null,3839,1085],"8939152cd98d1238362ec642fd6b6e350498f222ad8964388d53a5f67b4a0a2d",0]
+["tcgen05_mma_ws_tensor",0,0,"sm_110f","+ptx90","sm_110f",[3840,null,null,3841,1088],"34bebd948cff30e86671f16b3554603ee5e7eb4328f06decab48a07d51ecb8b3",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_100a","+ptx86","sm_100a",[3842,null,null,3843,213],"1c2f73750de72adf8cd1fc616e23bfff07cc99cb928a30575104efaf63e78b6e",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_100f","+ptx88","sm_100f",[3844,null,null,3845,1076],"fa341998cdd0f6ba7a28e68220abf433bf1b9f023c97d12bdf769b3f53ba5008",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_101a","+ptx86",null,[3846,null,null,3847,null],"9e4a176e6c6b829870d0241ef3b6f8f88f409218e04d876d2eda324894c82a86",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_101f","+ptx88",null,[3848,null,null,3849,null],"3d085e596d844f2c653d90de88ee2dd34790fb0aae52cfb46ee2d9a25cc042d5",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_103a","+ptx88","sm_103a",[3850,null,null,3851,1079],"a454ce9221176735a83373ff85a3f02fe7734e90341931a53877376cd4b2c7bb",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_103f","+ptx88","sm_103f",[3852,null,null,3853,1082],"050b3122d8df4d2b32d4a09d2a10988004830aa474301bb7396a777b5f198378",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_110a","+ptx90","sm_110a",[3854,null,null,3855,1085],"8de68a750461b7855540508dd1f12db93ba48f8a92fb5d490414e2c7216a46bb",0]
+["tcgen05_mma_ws_tensor_zero_col_mask",0,0,"sm_110f","+ptx90","sm_110f",[3856,null,null,3857,1088],"7603969ca011786f933cfd72c889ff4ff7528e232e11b90777ba21b1000c02b2",0]
+["tcgen05_mma_ws_tf32",0,0,"sm_100a","+ptx86","sm_100a",[3858,null,null,3859,213],"eb10a0976768470999856c5c58c0c04f1262bcdfe7f737705c82bfd767692aa8",0]
+["tcgen05_relinquish_alloc_permit",0,0,"sm_100a","+ptx86","sm_100a",[3860,null,null,3861,3217],"ab365a456117098087a1a192f8bd49b084a9633bf2975abdbb68791d5bcb669d",0]
+["tcgen05_relinquish_alloc_permit_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3862,null,null,3863,3217],"b1e25878a67f67cbb31f3ab8919e5dd89ce48ac8191afd4535789ab2566857e1",0]
+["tcgen05_shift_down",0,0,"sm_100a","+ptx86","sm_100a",[3864,null,null,3865,213],"0fba85f93d31303c84fcbeca4afa6de0639fc00aac3763b4373ed63a39b5ce85",0]
+["tcgen05_shift_down_cg2",0,0,"sm_100a","+ptx86","sm_100a",[3866,null,null,3867,213],"7e7b88e781a8349bcaad4c688ca4690ed431f811773c50870f3c516bae946e8d",0]
+["tcgen05_st_16x128b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3868,null,null,3869,213],"0803184178c5b9534c387e28b40ff60463b9143e93d6d0fc3f7b6261925b445c",0]
+["tcgen05_st_16x128b_x16_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3870,null,null,3871,213],"055db2897eecef9ef0c9e5a5b2a599a17efc06da7bf36f7b9ed3192893b7589e",0]
+["tcgen05_st_16x128b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3872,null,null,3873,213],"a3b5483df4d1678b5289438271de5cb31fd235d2d45244536b3bb690e1001794",0]
+["tcgen05_st_16x128b_x1_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3874,null,null,3875,213],"8cd97d8f99070321e35eb6b0c4d56039c079e2156f2b79306bdace54798b7b75",0]
+["tcgen05_st_16x128b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3876,null,null,3877,213],"41d9fa6f1aad68737858dd7bcec46da97c8242f02ac7adefac888a58f77c590a",0]
+["tcgen05_st_16x128b_x2_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3878,null,null,3879,213],"7f86656761247d7b5a6646b75556c48202b0c8acf913e7b645cb4f652d380b02",0]
+["tcgen05_st_16x128b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3880,null,null,3881,213],"585c31a1813195267b58c8c79a8e032de1f2c4dcb8ccec71a4648d03ba2f7963",0]
+["tcgen05_st_16x128b_x32_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3882,null,null,3883,213],"70b69bcb76120ce9770329ded1c0949430f4cd0cc05dab334d18363cd0fe2f50",0]
+["tcgen05_st_16x128b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3884,null,null,3885,213],"536be07a2d8672cba3bf7a7fb7332aec625b41f8e5f7488ca82cb30693f36a11",0]
+["tcgen05_st_16x128b_x4_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3886,null,null,3887,213],"61c673d45f1efece9e044db23d1cd61039757b0b612792d3c1a5728f11d06259",0]
+["tcgen05_st_16x128b_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3888,null,null,3889,213],"bbb5a56c42b40e88581e537bfc091d32b33af1baf18b00475b173bba62092275",0]
+["tcgen05_st_16x128b_x64_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3890,null,null,3891,213],"82a1fadaba1eff82077a13b6e2b19473c42cf319b17df60aa8453c9ac179cf61",0]
+["tcgen05_st_16x128b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3892,null,null,3893,213],"bbec1a7d95a0e73e7d8d753d0b68129e4322b1a12b8456321851c621a48825ff",0]
+["tcgen05_st_16x128b_x8_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3894,null,null,3895,213],"b2acae6097e2e6baa99eec478fc6a6df1d8ef31e14035915a71b9e12b01a8f1b",0]
+["tcgen05_st_16x256b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3896,null,null,3897,213],"e1961c68925f3c9cf23d04048ae245533119ce69c52e917b2c8f68794644e228",0]
+["tcgen05_st_16x256b_x16_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3898,null,null,3899,213],"dace153dd6b635e34acb98b575ef5123766ebb75d10122857bbe155acf17a622",0]
+["tcgen05_st_16x256b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3900,null,null,3901,213],"d82c4a6629a4d6aba31e30a7548794fc72654b36cc6271459e874e2fc625b062",0]
+["tcgen05_st_16x256b_x1_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3902,null,null,3903,213],"c7cef3d8f1107aa70fb8070a1f16c254e2819a1b242593475d1236bc46a6ef40",0]
+["tcgen05_st_16x256b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3904,null,null,3905,213],"8df0705a41318b92be00754e91acd9fe47a4385effef372eb4fdab0b3a95c8c6",0]
+["tcgen05_st_16x256b_x2_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3906,null,null,3907,213],"2d09b7847f667f0445aabad55ae4e8b8cbfdfa2517e0eda7fa7001bba3fa44e6",0]
+["tcgen05_st_16x256b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3908,null,null,3909,213],"20964203b6f71dfb034149e683a241d63da5420c7b779e404faac57521b9e005",0]
+["tcgen05_st_16x256b_x32_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3910,null,null,3911,213],"c5fa0368fe8a5c58182f8a28ac270fbcaaf267f8237ca1568490a42dc428847d",0]
+["tcgen05_st_16x256b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3912,null,null,3913,213],"c93a465ff6bd673934cc9450f2b6bddb2befd601d246bf4634251a0bb869b05a",0]
+["tcgen05_st_16x256b_x4_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3914,null,null,3915,213],"07c9eac6d2cafbc12ae4bf5c57534e3c7b3380f177254b45578e5540754c34cc",0]
+["tcgen05_st_16x256b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3916,null,null,3917,213],"d1289610f8fb9d9bbb0c70b67a4918055ade04565c75706c8740c04005dfc077",0]
+["tcgen05_st_16x256b_x8_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3918,null,null,3919,213],"54333cefa124f54e1089b0e7477e9d541ff0a33cc9e6404999ccfa0ca82b6c5a",0]
+["tcgen05_st_16x32bx2_x128_raw",0,0,"sm_100a","+ptx86","sm_100a",[3920,null,null,3921,213],"06838dfc75e374cc7bb1a451a9e30e57733896f51c21a1c30c5020a1da70af68",0]
+["tcgen05_st_16x32bx2_x128_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3922,null,null,3923,213],"ecbe93fbf93736bf88d521fb7094a47f779fa970c3a1279b193925d709dd508a",0]
+["tcgen05_st_16x32bx2_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3924,null,null,3925,213],"88839748938b5a498d72555ada9ac3ac4d508f6fbe178d7b8adb60354207f55a",0]
+["tcgen05_st_16x32bx2_x16_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3926,null,null,3927,213],"7b1e25bf33ea9a08684db3db7ed79020b78f9f4ececf25bf1595f788a82195a4",0]
+["tcgen05_st_16x32bx2_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3928,null,null,3929,213],"cec8b6783cab6771dad5fdffc7328d42ce05c646e1f64af0c6d8e14527e9d0a8",0]
+["tcgen05_st_16x32bx2_x1_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3930,null,null,3931,213],"f913325063a912d3afd22bdaa5fae142fc8d7acde2ea2e76234ac528bb13b03b",0]
+["tcgen05_st_16x32bx2_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3932,null,null,3933,213],"96a7a70424834d0e3df405e200c6591f53b83cf19b74349a2f7f7bc6b96cc711",0]
+["tcgen05_st_16x32bx2_x2_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3934,null,null,3935,213],"02020c0d9a34538c51d58d4807ea3de18e6e10b2f0ba9bcb839f1abb25cb8d10",0]
+["tcgen05_st_16x32bx2_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3936,null,null,3937,213],"4d8500760e4466c67df6681eb1adcaacb293fb8706f58d2414f6726917426834",0]
+["tcgen05_st_16x32bx2_x32_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3938,null,null,3939,213],"79d62e97bab7a4b46429ce48d2e304a04e290293b8f342ddf027dc9f3342f3ae",0]
+["tcgen05_st_16x32bx2_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3940,null,null,3941,213],"1fac86012ab10cafdc099a636ebf40ac737bb4f06b83955bbeb17fa029f96bb6",0]
+["tcgen05_st_16x32bx2_x4_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3942,null,null,3943,213],"a6778948bd3ac991a493a2d858359b56516a470cfd0bd846557268d918f12304",0]
+["tcgen05_st_16x32bx2_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3944,null,null,3945,213],"631db424e414d551c91d479bfd8d6a95c8a1828705e800ae4e28118487c1a019",0]
+["tcgen05_st_16x32bx2_x64_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3946,null,null,3947,213],"6db9e0a258783aba1d1ad96444de87f23b994cd1e4573ca538dc4684203c9214",0]
+["tcgen05_st_16x32bx2_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3948,null,null,3949,213],"dfe4fefcead0003c97de684232201d443cbf53f0ebb5f71a2662943ab068be80",0]
+["tcgen05_st_16x32bx2_x8_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3950,null,null,3951,213],"a9d4a87c952087c5e9d21b7531aa1aa96f3141568ee07df54c645ddf73eb38ad",0]
+["tcgen05_st_16x64b_x128_raw",0,0,"sm_100a","+ptx86","sm_100a",[3952,null,null,3953,213],"c6df7066a549248e7a8d0e74c9844c3f86f8fd7e4e3fc6a23d9eab3fe998d12e",0]
+["tcgen05_st_16x64b_x128_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3954,null,null,3955,213],"fbe29174e2eadeedb1f72998324db9226a3b851b8d8c81d210ada9028ec5337d",0]
+["tcgen05_st_16x64b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3956,null,null,3957,213],"4a1c47f70ed909872aa850ceb7a3eb3a5bd88c8bf9b6e09aad6e4bc2bec921f7",0]
+["tcgen05_st_16x64b_x16_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3958,null,null,3959,213],"69bc612ae7bb95a0de83e21f8a0717d75b749a589d1565015d8b3db9ac116054",0]
+["tcgen05_st_16x64b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3960,null,null,3961,213],"833659841b283ed69aff2cac7b1915ba93e93bc5e0d9859bcdf1c4537fd5b050",0]
+["tcgen05_st_16x64b_x1_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3962,null,null,3963,213],"437471c74d9e64e8b62331f4fe7b174e862014ee3f50d5c9dcaad05ae7a707c6",0]
+["tcgen05_st_16x64b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3964,null,null,3965,213],"5ecb17026e789b845cefcfad626b78d721f10d798bef1fa590753e2a1f6e2716",0]
+["tcgen05_st_16x64b_x2_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3966,null,null,3967,213],"747c6adaf321abe8ffe0b2fdd4d0c0a786e1fa38918d77c0b5568036053ad94f",0]
+["tcgen05_st_16x64b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[3968,null,null,3969,213],"855c27bd6efd38577be45be985211acb0ea1205ea108184d0b142782e577c124",0]
+["tcgen05_st_16x64b_x32_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3970,null,null,3971,213],"5c9bcb1d24d636e8916f66e05b4ff5032f5c23c8080a93e2e3e2e503e1aabfb6",0]
+["tcgen05_st_16x64b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[3972,null,null,3973,213],"310b9bdcb83f36345df805c6911d2e1b5eec40ba9962e0fff56e6a09665e69c5",0]
+["tcgen05_st_16x64b_x4_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3974,null,null,3975,213],"b6646479ae2eca7d526657921d5aa6e8ff5410432e247796f3d645461d572a46",0]
+["tcgen05_st_16x64b_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[3976,null,null,3977,213],"709c853130cd835f0ac80f75bd5ea733e083022726ea1d61401a59b0cd967cbb",0]
+["tcgen05_st_16x64b_x64_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3978,null,null,3979,213],"14fd968fc1e2ba486044b4b547ffd6dac2c66481335e3548aeaf7866c12cdb55",0]
+["tcgen05_st_16x64b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[3980,null,null,3981,213],"9a356bd66b998656bc29abebed531f8a0c8fe996d4b344e6787a601d35e808ba",0]
+["tcgen05_st_16x64b_x8_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3982,null,null,3983,213],"67c804b33e0ce63f419d6202c5441a275029595fad77c66e6f5d00bcc75bdd95",0]
+["tcgen05_st_32x32b_x128_raw",0,0,"sm_100a","+ptx86","sm_100a",[3984,null,null,3985,213],"21c25f7d41ca7b4045e86f280ca91b7113abfb262dd6c8811f20c4cab351ca67",0]
+["tcgen05_st_32x32b_x128_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3986,null,null,3987,213],"96f19d2417d7fa470420af6185c4be93e147bccd1db008d54cdb67603a10c7a7",0]
+["tcgen05_st_32x32b_x16_raw",0,0,"sm_100a","+ptx86","sm_100a",[3988,null,null,3989,213],"9d224208a4de18dc3a1423c9e388d8d36de4ea548f237f4dc0dced7080b3620e",0]
+["tcgen05_st_32x32b_x16_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3990,null,null,3991,213],"3f079dad2f77f912f2f82711aba20a4cb13792bdd8d9aebe7672b7dd35878c9f",0]
+["tcgen05_st_32x32b_x1_raw",0,0,"sm_100a","+ptx86","sm_100a",[3992,null,null,3993,213],"72cb12b288bb3ebe0545c65df1c77696f761226690075fdfc5a1c459b22a6509",0]
+["tcgen05_st_32x32b_x1_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3994,null,null,3995,213],"aaf485e77d93dbcbd8cd53eaf30f51533a88dc15c164ba926d75cf386c9dc41b",0]
+["tcgen05_st_32x32b_x2_raw",0,0,"sm_100a","+ptx86","sm_100a",[3996,null,null,3997,213],"ad0f15cc963e8715a0bb0274663aa6f14d63a51ffb50da598a37aafc54c1f03a",0]
+["tcgen05_st_32x32b_x2_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[3998,null,null,3999,213],"75c20604129f1fdc2d66b19b423f75c485a81249f9554244a78df735e6de04cc",0]
+["tcgen05_st_32x32b_x32_raw",0,0,"sm_100a","+ptx86","sm_100a",[4000,null,null,4001,213],"d2a695a3b90bc88d70bbd88cbcbf34f12aa6f07e429af02a93ee6b694aae6415",0]
+["tcgen05_st_32x32b_x32_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[4002,null,null,4003,213],"f3887a6e7cf419b3e6b176e26810b35bd27a78d4b764d898fc909b13fac66d49",0]
+["tcgen05_st_32x32b_x4_raw",0,0,"sm_100a","+ptx86","sm_100a",[4004,null,null,4005,213],"0de7a0a1f3f727eacfd69a952da89628d42fab86f408135026d4bc6ea6ee0088",0]
+["tcgen05_st_32x32b_x4_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[4006,null,null,4007,213],"b9408f7dcfec9fc9f5915e49df164f7756d01a44a9de99bdae21aaa370317dae",0]
+["tcgen05_st_32x32b_x64_raw",0,0,"sm_100a","+ptx86","sm_100a",[4008,null,null,4009,213],"87486ca12e6ab1a7206b3485642adccc1f1400ddfcfc61cff409dcda24a4c7a0",0]
+["tcgen05_st_32x32b_x64_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[4010,null,null,4011,213],"70aa28b8599f3ff579ab073119fe2c7181bc01064488baa1548a52be4903977a",0]
+["tcgen05_st_32x32b_x8_raw",0,0,"sm_100a","+ptx86","sm_100a",[4012,null,null,4013,213],"b8ae75638203357d6376933b203d2d1983dde2995512717c5efbc62e2e5df14d",0]
+["tcgen05_st_32x32b_x8_unpack16",0,0,"sm_100a","+ptx86","sm_100a",[4014,null,null,4015,213],"02a48e18f55a8a0568aa0ffa75b7694f99ed6f4427c66d3c7a5ee145de422d77",0]
+["tcgen05_store_wait",0,0,"sm_100a","+ptx86","sm_100a",[4016,null,null,4017,213],"86fb294947162282ba45a2067f9bb9a030fab2baef7492702aaff3554fec8371",0]
+["tensormap_replace_box_dim",0,0,"sm_90a","+ptx83","sm_90a",[4018,null,null,4019,4020],"4d7359f8e5937f0d37cc46442d97cffa4ec3017f31e7e17902167f402f368c6b",0]
+["tensormap_replace_element_stride",0,0,"sm_90a","+ptx83","sm_90a",[4021,null,null,4022,4020],"0e85fffa1842a7542054a827c69aef331dab6195ee99208792d1b8b05d241f44",0]
+["tensormap_replace_element_type",0,0,"sm_90a","+ptx83","sm_90a",[4023,null,null,4024,4020],"34f3224d49e8b3abaeebf8704a896c3693c7319d9bdeca2c7d149c61e8bde8a9",0]
+["tensormap_replace_fill_mode",0,0,"sm_90a","+ptx83","sm_90a",[4025,null,null,4026,4020],"60c1147d9cfd178f7048a1a8457647d90e422d031d280bda995ad22f0c0d8a4b",0]
+["tensormap_replace_global_address",0,0,"sm_90a","+ptx83","sm_90a",[4027,null,null,4028,4020],"92e38cb091ea2e3380d5b12781a2e8269c84f235b8747111a76caf08624243f7",0]
+["tensormap_replace_global_dim",0,0,"sm_90a","+ptx83","sm_90a",[4029,null,null,4030,4020],"6e257beba486f8a27ec223ea4fcfd357a4ecf279a5b1969e51364a4348fdd89d",0]
+["tensormap_replace_global_stride",0,0,"sm_90a","+ptx83","sm_90a",[4031,null,null,4032,4020],"5b4ab15a2e839b3d7a6dc458be2ebdaebd5c6e644d46a7cea6f163c466744c28",0]
+["tensormap_replace_interleave_layout",0,0,"sm_90a","+ptx83","sm_90a",[4033,null,null,4034,4020],"ff05b77d28f1a70f0559bc80dbbfc1623494c4a36ea9aae2db70a911f679b4c1",0]
+["tensormap_replace_rank",0,0,"sm_90a","+ptx83","sm_90a",[4035,null,null,4036,4020],"9e01d000bd520d83932828ac524fdd9facf99eb920c7a94abf59ab16aff25d95",0]
+["tensormap_replace_swizzle_atomicity",0,0,"sm_100a","+ptx86","sm_100a",[4037,null,null,4038,4039],"79cceb41f28df866bc6dd37411ef6accb74cd748aa0cfa09811da47647f4aa7d",0]
+["tensormap_replace_swizzle_mode",0,0,"sm_90a","+ptx83","sm_90a",[4040,null,null,4041,4020],"9cd96f713d701616cd76a62c3423998d98db18498d5a79941f50170cbea57c84",0]
+["thread_idx_x",0,0,"sm_70","+ptx60",null,[4042,4043,4044,4045,null],"0146a8b340aeab76891aea97ac6cb5ff6dd26bc22cf27f95405c588496802ec8",0]
+["thread_idx_y",0,0,"sm_70","+ptx60",null,[4046,4047,4048,4049,null],"1e2749b85469030f50fed0223fefc4054c02ab5e626be0287ea516744034b296",0]
+["thread_idx_z",0,0,"sm_70","+ptx60",null,[4050,4051,4052,4053,null],"eaaa9912ea88309e88536f5c3a9233eedc61021c9d6d0ea7c4b248e4ab00eba2",0]
+["threadfence",0,0,"sm_20","+ptx32",null,[4054,4055,4056,4057,null],"f17f219c9a2b9a27feb4e1832cadbc69180d146505999cf185959a107b6c59ac",0]
+["threadfence",1,0,"sm_20","+ptx32","sm_75",[4058,4060,4059,4057,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["threadfence_block",0,0,"sm_20","+ptx32",null,[4061,4062,4063,4064,null],"ac69a8352f3af3d8e163a2f34557d83d813d830c8f0d8cb84a5f10b25b313f6e",0]
+["threadfence_block",1,0,"sm_20","+ptx32","sm_75",[4065,4067,4066,4064,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["threadfence_system",0,0,"sm_20","+ptx32",null,[4068,4069,4070,4071,null],"1e8257694b9c25f72d1153fe27b7311c3ea3c9bb0e47b7b53ff7c4a9035380a4",0]
+["threadfence_system",1,0,"sm_20","+ptx32","sm_75",[4072,4074,4073,4071,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["total_smem_size",1,0,"sm_20","+ptx41",null,[4075,null,null,4076,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["total_smem_size",1,0,"sm_20","+ptx41","sm_75",[4075,null,null,4076,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["trap",0,0,"sm_20","+ptx32",null,[4077,null,null,4078,null],"640986c81812457a2e4b8f60df66e154a72af15d84a319b52326929129ff60ba",0]
+["trap",1,0,"sm_20","+ptx32","sm_75",[4079,null,null,4078,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",1]
+["uni_sync",0,0,"sm_30","+ptx60",null,[4080,4081,4082,4083,null],"c5f7cd5264eff84c49bf43cf9c990e771781e0ae180404c7c3220222af6fa7d2",0]
+["warpid",1,0,"sm_20","+ptx32",null,[4084,null,null,4085,null],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["warpid",1,0,"sm_20","+ptx32","sm_75",[4084,null,null,4085,146],"016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",101]
+["wgmma_commit_group",0,0,"sm_90a","+ptx80","sm_90a",[4086,4087,4088,4089,3039],"693c6befba457f477670a27edd4d121467b23334a61bbc97fb4bd948df82c2d3",0]
+["wgmma_fence",0,0,"sm_90a","+ptx80","sm_90a",[4090,4091,4092,4093,3039],"287dcef184688aa23edfdd45c54c4b619fac08736e1d398a626579c7ac8bac2b",0]
+["wgmma_wait_group",0,0,"sm_90a","+ptx80","sm_90a",[4094,4095,4096,4097,3039],"8f10b0161b99a898d44c8d5c73bb7a03320b9faf6cbd9e1ea13c63d69c58ba21",0]
+["add_rm_ftz_sat_f32",2,1,null,null,null,[4098,4099,4100,null,null],null,0]
+["add_rm_sat_f32",2,1,null,null,null,[4101,4102,4103,null,null],null,0]
+["add_rn_ftz_sat_f32",2,1,null,null,null,[4104,4105,4106,null,null],null,0]
+["add_rn_sat_f32",2,1,null,null,null,[4107,4108,4109,null,null],null,0]
+["add_rp_ftz_sat_f32",2,1,null,null,null,[4110,4111,4112,null,null],null,0]
+["add_rp_sat_f32",2,1,null,null,null,[4113,4114,4115,null,null],null,0]
+["add_rz_ftz_sat_f32",2,1,null,null,null,[4116,4117,4118,null,null],null,0]
+["add_rz_sat_f32",2,1,null,null,null,[4119,4120,4121,null,null],null,0]
+["cos_approx_f32",2,1,null,null,null,[4122,4123,4124,null,null],null,0]
+["cos_approx_ftz_f32",2,1,null,null,null,[4125,4126,4127,null,null],null,0]
+["cvt_f16x2_f32",2,1,null,null,null,[4128,4129,4130,null,null],null,0]
+["cvt_f16x2_f32",2,2,"sm_80","+ptx70",null,[4131,null,null,4132,null],null,0]
+["cvt_f16x2_f32",2,3,"sm_80","+ptx70","sm_80",[4131,null,null,4132,2],null,0]
+["cvt_f32x2_bf16x2",2,1,null,null,null,[4133,4134,4135,null,null],null,0]
+["cvt_f32x2_bf16x2",2,2,"sm_80","+ptx70",null,[4136,null,null,4137,null],null,0]
+["cvt_f32x2_bf16x2",2,3,"sm_80","+ptx70","sm_80",[4136,null,null,4137,2],null,0]
+["cvt_rn_f16x2_e4m3x2",2,1,null,null,null,[4138,4139,4140,null,null],null,0]
+["cvt_rn_f16x2_e4m3x2",2,2,"sm_89","+ptx81",null,[4141,null,null,4142,null],null,0]
+["cvt_rn_f16x2_e4m3x2",2,3,"sm_89","+ptx81","sm_89",[4141,null,null,4142,714],null,0]
+["cvt_rn_f16x2_e5m2x2",2,1,null,null,null,[4143,4144,4145,null,null],null,0]
+["cvt_rn_f16x2_e5m2x2",2,2,"sm_89","+ptx81",null,[4146,null,null,4147,null],null,0]
+["cvt_rn_f16x2_e5m2x2",2,3,"sm_89","+ptx81","sm_89",[4146,null,null,4147,714],null,0]
+["cvt_rn_relu_bf16x2_f32",2,1,null,null,null,[4148,4149,4150,null,null],null,0]
+["cvt_rn_relu_bf16x2_f32",2,2,"sm_80","+ptx70",null,[4151,null,null,4152,null],null,0]
+["cvt_rn_relu_bf16x2_f32",2,3,"sm_80","+ptx70","sm_80",[4151,null,null,4152,2],null,0]
+["cvt_rn_relu_f16x2_e4m3x2",2,1,null,null,null,[4153,4154,4155,null,null],null,0]
+["cvt_rn_relu_f16x2_e4m3x2",2,2,"sm_89","+ptx81",null,[4156,null,null,4157,null],null,0]
+["cvt_rn_relu_f16x2_e4m3x2",2,3,"sm_89","+ptx81","sm_89",[4156,null,null,4157,714],null,0]
+["cvt_rn_relu_f16x2_e5m2x2",2,1,null,null,null,[4158,4159,4160,null,null],null,0]
+["cvt_rn_relu_f16x2_e5m2x2",2,2,"sm_89","+ptx81",null,[4161,null,null,4162,null],null,0]
+["cvt_rn_relu_f16x2_e5m2x2",2,3,"sm_89","+ptx81","sm_89",[4161,null,null,4162,714],null,0]
+["cvt_rn_relu_f16x2_f32",2,1,null,null,null,[4163,4164,4165,null,null],null,0]
+["cvt_rn_relu_f16x2_f32",2,2,"sm_80","+ptx70",null,[4166,null,null,4167,null],null,0]
+["cvt_rn_relu_f16x2_f32",2,3,"sm_80","+ptx70","sm_80",[4166,null,null,4167,2],null,0]
+["cvt_rn_satfinite_e4m3x2_f16x2",2,1,null,null,null,[4168,4169,4170,null,null],null,0]
+["cvt_rn_satfinite_e4m3x2_f16x2",2,2,"sm_89","+ptx81",null,[4171,null,null,4172,null],null,0]
+["cvt_rn_satfinite_e4m3x2_f16x2",2,3,"sm_89","+ptx81","sm_89",[4171,null,null,4172,714],null,0]
+["cvt_rn_satfinite_e5m2x2_f16x2",2,1,null,null,null,[4173,4174,4175,null,null],null,0]
+["cvt_rn_satfinite_e5m2x2_f16x2",2,2,"sm_89","+ptx81",null,[4176,null,null,4177,null],null,0]
+["cvt_rn_satfinite_e5m2x2_f16x2",2,3,"sm_89","+ptx81","sm_89",[4176,null,null,4177,714],null,0]
+["cvt_rn_satfinite_relu_e4m3x2_f16x2",2,1,null,null,null,[4178,4179,4180,null,null],null,0]
+["cvt_rn_satfinite_relu_e4m3x2_f16x2",2,2,"sm_89","+ptx81",null,[4181,null,null,4182,null],null,0]
+["cvt_rn_satfinite_relu_e4m3x2_f16x2",2,3,"sm_89","+ptx81","sm_89",[4181,null,null,4182,714],null,0]
+["cvt_rn_satfinite_relu_e5m2x2_f16x2",2,1,null,null,null,[4183,4184,4185,null,null],null,0]
+["cvt_rn_satfinite_relu_e5m2x2_f16x2",2,2,"sm_89","+ptx81",null,[4186,null,null,4187,null],null,0]
+["cvt_rn_satfinite_relu_e5m2x2_f16x2",2,3,"sm_89","+ptx81","sm_89",[4186,null,null,4187,714],null,0]
+["cvt_rz_bf16x2_f32",2,1,null,null,null,[4188,4189,4190,null,null],null,0]
+["cvt_rz_bf16x2_f32",2,2,"sm_80","+ptx70",null,[4191,null,null,4192,null],null,0]
+["cvt_rz_bf16x2_f32",2,3,"sm_80","+ptx70","sm_80",[4191,null,null,4192,2],null,0]
+["cvt_rz_f16x2_f32",2,1,null,null,null,[4193,4194,4195,null,null],null,0]
+["cvt_rz_f16x2_f32",2,2,"sm_80","+ptx70",null,[4196,null,null,4197,null],null,0]
+["cvt_rz_f16x2_f32",2,3,"sm_80","+ptx70","sm_80",[4196,null,null,4197,2],null,0]
+["fma_rm_ftz_sat_f32",2,1,null,null,null,[4198,4199,4200,null,null],null,0]
+["fma_rm_sat_f32",2,1,null,null,null,[4201,4202,4203,null,null],null,0]
+["fma_rn_ftz_sat_f32",2,1,null,null,null,[4204,4205,4206,null,null],null,0]
+["fma_rn_sat_f32",2,1,null,null,null,[4207,4208,4209,null,null],null,0]
+["fma_rp_ftz_sat_f32",2,1,null,null,null,[4210,4211,4212,null,null],null,0]
+["fma_rp_sat_f32",2,1,null,null,null,[4213,4214,4215,null,null],null,0]
+["fma_rz_ftz_sat_f32",2,1,null,null,null,[4216,4217,4218,null,null],null,0]
+["fma_rz_sat_f32",2,1,null,null,null,[4219,4220,4221,null,null],null,0]
+["lg2_approx_f32",2,1,null,null,null,[4222,4223,4224,null,null],null,0]
+["lg2_approx_ftz_f32",2,1,null,null,null,[4225,4226,4227,null,null],null,0]
+["max_bf16",2,1,null,null,null,[4228,4229,4230,null,null],null,0]
+["max_bf16",2,2,"sm_80","+ptx70",null,[4231,null,null,4232,null],null,0]
+["max_bf16",2,3,"sm_80","+ptx70","sm_80",[4231,null,null,4232,2],null,0]
+["max_f16",2,1,null,null,null,[4233,4234,4235,null,null],null,0]
+["max_f16",2,2,"sm_80","+ptx70",null,[4236,null,null,4237,null],null,0]
+["max_f16",2,3,"sm_80","+ptx70","sm_80",[4236,null,null,4237,2],null,0]
+["max_ftz_f16",2,1,null,null,null,[4238,4239,4240,null,null],null,0]
+["max_ftz_f16",2,2,"sm_80","+ptx70",null,[4241,null,null,4242,null],null,0]
+["max_ftz_f16",2,3,"sm_80","+ptx70","sm_80",[4241,null,null,4242,2],null,0]
+["max_ftz_f16x2",2,1,null,null,null,[4243,4244,4245,null,null],null,0]
+["max_ftz_f16x2",2,2,"sm_80","+ptx70",null,[4246,null,null,4247,null],null,0]
+["max_ftz_f16x2",2,3,"sm_80","+ptx70","sm_80",[4246,null,null,4247,2],null,0]
+["max_ftz_nan_f16",2,1,null,null,null,[4248,4249,4250,null,null],null,0]
+["max_ftz_nan_f16",2,2,"sm_80","+ptx70",null,[4251,null,null,4252,null],null,0]
+["max_ftz_nan_f16",2,3,"sm_80","+ptx70","sm_80",[4251,null,null,4252,2],null,0]
+["max_ftz_nan_f16x2",2,1,null,null,null,[4253,4254,4255,null,null],null,0]
+["max_ftz_nan_f16x2",2,2,"sm_80","+ptx70",null,[4256,null,null,4257,null],null,0]
+["max_ftz_nan_f16x2",2,3,"sm_80","+ptx70","sm_80",[4256,null,null,4257,2],null,0]
+["max_ftz_nan_xorsign_abs_f16",2,1,null,null,null,[4258,4259,4260,null,null],null,0]
+["max_ftz_nan_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4261,null,null,4262,null],null,0]
+["max_ftz_nan_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4261,null,null,4262,1407],null,0]
+["max_ftz_nan_xorsign_abs_f16x2",2,1,null,null,null,[4263,4264,4265,null,null],null,0]
+["max_ftz_nan_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4266,null,null,4267,null],null,0]
+["max_ftz_nan_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4266,null,null,4267,1407],null,0]
+["max_ftz_nan_xorsign_abs_f32",2,1,null,null,null,[4268,4269,4270,null,null],null,0]
+["max_ftz_nan_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4271,null,null,4272,null],null,0]
+["max_ftz_nan_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4271,null,null,4272,1407],null,0]
+["max_ftz_xorsign_abs_f16",2,1,null,null,null,[4273,4274,4275,null,null],null,0]
+["max_ftz_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4276,null,null,4277,null],null,0]
+["max_ftz_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4276,null,null,4277,1407],null,0]
+["max_ftz_xorsign_abs_f16x2",2,1,null,null,null,[4278,4279,4280,null,null],null,0]
+["max_ftz_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4281,null,null,4282,null],null,0]
+["max_ftz_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4281,null,null,4282,1407],null,0]
+["max_ftz_xorsign_abs_f32",2,1,null,null,null,[4283,4284,4285,null,null],null,0]
+["max_ftz_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4286,null,null,4287,null],null,0]
+["max_ftz_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4286,null,null,4287,1407],null,0]
+["max_nan_bf16",2,1,null,null,null,[4288,4289,4290,null,null],null,0]
+["max_nan_bf16",2,2,"sm_80","+ptx70",null,[4291,null,null,4292,null],null,0]
+["max_nan_bf16",2,3,"sm_80","+ptx70","sm_80",[4291,null,null,4292,2],null,0]
+["max_nan_bf16x2",2,1,null,null,null,[4293,4294,4295,null,null],null,0]
+["max_nan_bf16x2",2,2,"sm_80","+ptx70",null,[4296,null,null,4297,null],null,0]
+["max_nan_bf16x2",2,3,"sm_80","+ptx70","sm_80",[4296,null,null,4297,2],null,0]
+["max_nan_f16",2,1,null,null,null,[4298,4299,4300,null,null],null,0]
+["max_nan_f16",2,2,"sm_80","+ptx70",null,[4301,null,null,4302,null],null,0]
+["max_nan_f16",2,3,"sm_80","+ptx70","sm_80",[4301,null,null,4302,2],null,0]
+["max_nan_f16x2",2,1,null,null,null,[4303,4304,4305,null,null],null,0]
+["max_nan_f16x2",2,2,"sm_80","+ptx70",null,[4306,null,null,4307,null],null,0]
+["max_nan_f16x2",2,3,"sm_80","+ptx70","sm_80",[4306,null,null,4307,2],null,0]
+["max_nan_xorsign_abs_bf16",2,1,null,null,null,[4308,4309,4310,null,null],null,0]
+["max_nan_xorsign_abs_bf16",2,2,"sm_86","+ptx72",null,[4311,null,null,4312,null],null,0]
+["max_nan_xorsign_abs_bf16",2,3,"sm_86","+ptx72","sm_86",[4311,null,null,4312,1407],null,0]
+["max_nan_xorsign_abs_bf16x2",2,1,null,null,null,[4313,4314,4315,null,null],null,0]
+["max_nan_xorsign_abs_bf16x2",2,2,"sm_86","+ptx78",null,[4316,null,null,4317,null],null,0]
+["max_nan_xorsign_abs_bf16x2",2,3,"sm_86","+ptx78","sm_86",[4316,null,null,4317,1407],null,0]
+["max_nan_xorsign_abs_f16",2,1,null,null,null,[4318,4319,4320,null,null],null,0]
+["max_nan_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4321,null,null,4322,null],null,0]
+["max_nan_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4321,null,null,4322,1407],null,0]
+["max_nan_xorsign_abs_f16x2",2,1,null,null,null,[4323,4324,4325,null,null],null,0]
+["max_nan_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4326,null,null,4327,null],null,0]
+["max_nan_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4326,null,null,4327,1407],null,0]
+["max_nan_xorsign_abs_f32",2,1,null,null,null,[4328,4329,4330,null,null],null,0]
+["max_nan_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4331,null,null,4332,null],null,0]
+["max_nan_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4331,null,null,4332,1407],null,0]
+["max_xorsign_abs_bf16",2,1,null,null,null,[4333,4334,4335,null,null],null,0]
+["max_xorsign_abs_bf16",2,2,"sm_86","+ptx72",null,[4336,null,null,4337,null],null,0]
+["max_xorsign_abs_bf16",2,3,"sm_86","+ptx72","sm_86",[4336,null,null,4337,1407],null,0]
+["max_xorsign_abs_bf16x2",2,1,null,null,null,[4338,4339,4340,null,null],null,0]
+["max_xorsign_abs_bf16x2",2,2,"sm_86","+ptx78",null,[4341,null,null,4342,null],null,0]
+["max_xorsign_abs_bf16x2",2,3,"sm_86","+ptx78","sm_86",[4341,null,null,4342,1407],null,0]
+["max_xorsign_abs_f16",2,1,null,null,null,[4343,4344,4345,null,null],null,0]
+["max_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4346,null,null,4347,null],null,0]
+["max_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4346,null,null,4347,1407],null,0]
+["max_xorsign_abs_f16x2",2,1,null,null,null,[4348,4349,4350,null,null],null,0]
+["max_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4351,null,null,4352,null],null,0]
+["max_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4351,null,null,4352,1407],null,0]
+["max_xorsign_abs_f32",2,1,null,null,null,[4353,4354,4355,null,null],null,0]
+["max_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4356,null,null,4357,null],null,0]
+["max_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4356,null,null,4357,1407],null,0]
+["min_bf16",2,1,null,null,null,[4358,4359,4360,null,null],null,0]
+["min_bf16",2,2,"sm_80","+ptx70",null,[4361,null,null,4362,null],null,0]
+["min_bf16",2,3,"sm_80","+ptx70","sm_80",[4361,null,null,4362,2],null,0]
+["min_f16",2,1,null,null,null,[4363,4364,4365,null,null],null,0]
+["min_f16",2,2,"sm_80","+ptx70",null,[4366,null,null,4367,null],null,0]
+["min_f16",2,3,"sm_80","+ptx70","sm_80",[4366,null,null,4367,2],null,0]
+["min_ftz_f16",2,1,null,null,null,[4368,4369,4370,null,null],null,0]
+["min_ftz_f16",2,2,"sm_80","+ptx70",null,[4371,null,null,4372,null],null,0]
+["min_ftz_f16",2,3,"sm_80","+ptx70","sm_80",[4371,null,null,4372,2],null,0]
+["min_ftz_f16x2",2,1,null,null,null,[4373,4374,4375,null,null],null,0]
+["min_ftz_f16x2",2,2,"sm_80","+ptx70",null,[4376,null,null,4377,null],null,0]
+["min_ftz_f16x2",2,3,"sm_80","+ptx70","sm_80",[4376,null,null,4377,2],null,0]
+["min_ftz_nan_f16",2,1,null,null,null,[4378,4379,4380,null,null],null,0]
+["min_ftz_nan_f16",2,2,"sm_80","+ptx70",null,[4381,null,null,4382,null],null,0]
+["min_ftz_nan_f16",2,3,"sm_80","+ptx70","sm_80",[4381,null,null,4382,2],null,0]
+["min_ftz_nan_f16x2",2,1,null,null,null,[4383,4384,4385,null,null],null,0]
+["min_ftz_nan_f16x2",2,2,"sm_80","+ptx70",null,[4386,null,null,4387,null],null,0]
+["min_ftz_nan_f16x2",2,3,"sm_80","+ptx70","sm_80",[4386,null,null,4387,2],null,0]
+["min_ftz_nan_xorsign_abs_f16",2,1,null,null,null,[4388,4389,4390,null,null],null,0]
+["min_ftz_nan_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4391,null,null,4392,null],null,0]
+["min_ftz_nan_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4391,null,null,4392,1407],null,0]
+["min_ftz_nan_xorsign_abs_f16x2",2,1,null,null,null,[4393,4394,4395,null,null],null,0]
+["min_ftz_nan_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4396,null,null,4397,null],null,0]
+["min_ftz_nan_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4396,null,null,4397,1407],null,0]
+["min_ftz_nan_xorsign_abs_f32",2,1,null,null,null,[4398,4399,4400,null,null],null,0]
+["min_ftz_nan_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4401,null,null,4402,null],null,0]
+["min_ftz_nan_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4401,null,null,4402,1407],null,0]
+["min_ftz_xorsign_abs_f16",2,1,null,null,null,[4403,4404,4405,null,null],null,0]
+["min_ftz_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4406,null,null,4407,null],null,0]
+["min_ftz_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4406,null,null,4407,1407],null,0]
+["min_ftz_xorsign_abs_f16x2",2,1,null,null,null,[4408,4409,4410,null,null],null,0]
+["min_ftz_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4411,null,null,4412,null],null,0]
+["min_ftz_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4411,null,null,4412,1407],null,0]
+["min_ftz_xorsign_abs_f32",2,1,null,null,null,[4413,4414,4415,null,null],null,0]
+["min_ftz_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4416,null,null,4417,null],null,0]
+["min_ftz_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4416,null,null,4417,1407],null,0]
+["min_nan_bf16",2,1,null,null,null,[4418,4419,4420,null,null],null,0]
+["min_nan_bf16",2,2,"sm_80","+ptx70",null,[4421,null,null,4422,null],null,0]
+["min_nan_bf16",2,3,"sm_80","+ptx70","sm_80",[4421,null,null,4422,2],null,0]
+["min_nan_bf16x2",2,1,null,null,null,[4423,4424,4425,null,null],null,0]
+["min_nan_bf16x2",2,2,"sm_80","+ptx70",null,[4426,null,null,4427,null],null,0]
+["min_nan_bf16x2",2,3,"sm_80","+ptx70","sm_80",[4426,null,null,4427,2],null,0]
+["min_nan_f16",2,1,null,null,null,[4428,4429,4430,null,null],null,0]
+["min_nan_f16",2,2,"sm_80","+ptx70",null,[4431,null,null,4432,null],null,0]
+["min_nan_f16",2,3,"sm_80","+ptx70","sm_80",[4431,null,null,4432,2],null,0]
+["min_nan_f16x2",2,1,null,null,null,[4433,4434,4435,null,null],null,0]
+["min_nan_f16x2",2,2,"sm_80","+ptx70",null,[4436,null,null,4437,null],null,0]
+["min_nan_f16x2",2,3,"sm_80","+ptx70","sm_80",[4436,null,null,4437,2],null,0]
+["min_nan_xorsign_abs_bf16",2,1,null,null,null,[4438,4439,4440,null,null],null,0]
+["min_nan_xorsign_abs_bf16",2,2,"sm_86","+ptx72",null,[4441,null,null,4442,null],null,0]
+["min_nan_xorsign_abs_bf16",2,3,"sm_86","+ptx72","sm_86",[4441,null,null,4442,1407],null,0]
+["min_nan_xorsign_abs_bf16x2",2,1,null,null,null,[4443,4444,4445,null,null],null,0]
+["min_nan_xorsign_abs_bf16x2",2,2,"sm_86","+ptx78",null,[4446,null,null,4447,null],null,0]
+["min_nan_xorsign_abs_bf16x2",2,3,"sm_86","+ptx78","sm_86",[4446,null,null,4447,1407],null,0]
+["min_nan_xorsign_abs_f16",2,1,null,null,null,[4448,4449,4450,null,null],null,0]
+["min_nan_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4451,null,null,4452,null],null,0]
+["min_nan_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4451,null,null,4452,1407],null,0]
+["min_nan_xorsign_abs_f16x2",2,1,null,null,null,[4453,4454,4455,null,null],null,0]
+["min_nan_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4456,null,null,4457,null],null,0]
+["min_nan_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4456,null,null,4457,1407],null,0]
+["min_nan_xorsign_abs_f32",2,1,null,null,null,[4458,4459,4460,null,null],null,0]
+["min_nan_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4461,null,null,4462,null],null,0]
+["min_nan_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4461,null,null,4462,1407],null,0]
+["min_xorsign_abs_bf16",2,1,null,null,null,[4463,4464,4465,null,null],null,0]
+["min_xorsign_abs_bf16",2,2,"sm_86","+ptx72",null,[4466,null,null,4467,null],null,0]
+["min_xorsign_abs_bf16",2,3,"sm_86","+ptx72","sm_86",[4466,null,null,4467,1407],null,0]
+["min_xorsign_abs_bf16x2",2,1,null,null,null,[4468,4469,4470,null,null],null,0]
+["min_xorsign_abs_bf16x2",2,2,"sm_86","+ptx78",null,[4471,null,null,4472,null],null,0]
+["min_xorsign_abs_bf16x2",2,3,"sm_86","+ptx78","sm_86",[4471,null,null,4472,1407],null,0]
+["min_xorsign_abs_f16",2,1,null,null,null,[4473,4474,4475,null,null],null,0]
+["min_xorsign_abs_f16",2,2,"sm_86","+ptx72",null,[4476,null,null,4477,null],null,0]
+["min_xorsign_abs_f16",2,3,"sm_86","+ptx72","sm_86",[4476,null,null,4477,1407],null,0]
+["min_xorsign_abs_f16x2",2,1,null,null,null,[4478,4479,4480,null,null],null,0]
+["min_xorsign_abs_f16x2",2,2,"sm_86","+ptx78",null,[4481,null,null,4482,null],null,0]
+["min_xorsign_abs_f16x2",2,3,"sm_86","+ptx78","sm_86",[4481,null,null,4482,1407],null,0]
+["min_xorsign_abs_f32",2,1,null,null,null,[4483,4484,4485,null,null],null,0]
+["min_xorsign_abs_f32",2,2,"sm_86","+ptx78",null,[4486,null,null,4487,null],null,0]
+["min_xorsign_abs_f32",2,3,"sm_86","+ptx78","sm_86",[4486,null,null,4487,1407],null,0]
+["mma_m16n8k16_f16_f16",2,1,null,null,null,[4488,4489,4490,null,null],null,0]
+["mma_m16n8k16_fp8_f16_e4m3_e4m3",2,1,null,null,null,[4491,4492,4493,null,null],null,0]
+["mma_m16n8k16_fp8_f16_e4m3_e5m2",2,1,null,null,null,[4494,4495,4496,null,null],null,0]
+["mma_m16n8k16_fp8_f16_e5m2_e4m3",2,1,null,null,null,[4497,4498,4499,null,null],null,0]
+["mma_m16n8k16_fp8_f16_e5m2_e5m2",2,1,null,null,null,[4500,4501,4502,null,null],null,0]
+["mma_m16n8k16_fp8_f32_e4m3_e4m3",2,1,null,null,null,[4503,4504,4505,null,null],null,0]
+["mma_m16n8k16_fp8_f32_e4m3_e5m2",2,1,null,null,null,[4506,4507,4508,null,null],null,0]
+["mma_m16n8k16_fp8_f32_e5m2_e4m3",2,1,null,null,null,[4509,4510,4511,null,null],null,0]
+["mma_m16n8k16_fp8_f32_e5m2_e5m2",2,1,null,null,null,[4512,4513,4514,null,null],null,0]
+["mma_m16n8k32_f16_e2m1_e2m1",2,1,null,null,null,[4515,4516,4517,null,null],null,0]
+["mma_m16n8k32_f16_e2m1_e2m1",2,2,"sm_120a","+ptx87",null,[4518,null,null,4519,null],null,0]
+["mma_m16n8k32_f16_e2m1_e2m3",2,1,null,null,null,[4520,4521,4522,null,null],null,0]
+["mma_m16n8k32_f16_e2m1_e2m3",2,2,"sm_120a","+ptx87",null,[4523,null,null,4524,null],null,0]
+["mma_m16n8k32_f16_e2m1_e3m2",2,1,null,null,null,[4525,4526,4527,null,null],null,0]
+["mma_m16n8k32_f16_e2m1_e3m2",2,2,"sm_120a","+ptx87",null,[4528,null,null,4529,null],null,0]
+["mma_m16n8k32_f16_e2m1_e4m3",2,1,null,null,null,[4530,4531,4532,null,null],null,0]
+["mma_m16n8k32_f16_e2m1_e4m3",2,2,"sm_120a","+ptx87",null,[4533,null,null,4534,null],null,0]
+["mma_m16n8k32_f16_e2m1_e5m2",2,1,null,null,null,[4535,4536,4537,null,null],null,0]
+["mma_m16n8k32_f16_e2m1_e5m2",2,2,"sm_120a","+ptx87",null,[4538,null,null,4539,null],null,0]
+["mma_m16n8k32_f16_e2m3_e2m1",2,1,null,null,null,[4540,4541,4542,null,null],null,0]
+["mma_m16n8k32_f16_e2m3_e2m1",2,2,"sm_120a","+ptx87",null,[4543,null,null,4544,null],null,0]
+["mma_m16n8k32_f16_e2m3_e2m3",2,1,null,null,null,[4545,4546,4547,null,null],null,0]
+["mma_m16n8k32_f16_e2m3_e2m3",2,2,"sm_120a","+ptx87",null,[4548,null,null,4549,null],null,0]
+["mma_m16n8k32_f16_e2m3_e3m2",2,1,null,null,null,[4550,4551,4552,null,null],null,0]
+["mma_m16n8k32_f16_e2m3_e3m2",2,2,"sm_120a","+ptx87",null,[4553,null,null,4554,null],null,0]
+["mma_m16n8k32_f16_e2m3_e4m3",2,1,null,null,null,[4555,4556,4557,null,null],null,0]
+["mma_m16n8k32_f16_e2m3_e4m3",2,2,"sm_120a","+ptx87",null,[4558,null,null,4559,null],null,0]
+["mma_m16n8k32_f16_e2m3_e5m2",2,1,null,null,null,[4560,4561,4562,null,null],null,0]
+["mma_m16n8k32_f16_e2m3_e5m2",2,2,"sm_120a","+ptx87",null,[4563,null,null,4564,null],null,0]
+["mma_m16n8k32_f16_e3m2_e2m1",2,1,null,null,null,[4565,4566,4567,null,null],null,0]
+["mma_m16n8k32_f16_e3m2_e2m1",2,2,"sm_120a","+ptx87",null,[4568,null,null,4569,null],null,0]
+["mma_m16n8k32_f16_e3m2_e2m3",2,1,null,null,null,[4570,4571,4572,null,null],null,0]
+["mma_m16n8k32_f16_e3m2_e2m3",2,2,"sm_120a","+ptx87",null,[4573,null,null,4574,null],null,0]
+["mma_m16n8k32_f16_e3m2_e3m2",2,1,null,null,null,[4575,4576,4577,null,null],null,0]
+["mma_m16n8k32_f16_e3m2_e3m2",2,2,"sm_120a","+ptx87",null,[4578,null,null,4579,null],null,0]
+["mma_m16n8k32_f16_e3m2_e4m3",2,1,null,null,null,[4580,4581,4582,null,null],null,0]
+["mma_m16n8k32_f16_e3m2_e4m3",2,2,"sm_120a","+ptx87",null,[4583,null,null,4584,null],null,0]
+["mma_m16n8k32_f16_e3m2_e5m2",2,1,null,null,null,[4585,4586,4587,null,null],null,0]
+["mma_m16n8k32_f16_e3m2_e5m2",2,2,"sm_120a","+ptx87",null,[4588,null,null,4589,null],null,0]
+["mma_m16n8k32_f16_e4m3_e2m1",2,1,null,null,null,[4590,4591,4592,null,null],null,0]
+["mma_m16n8k32_f16_e4m3_e2m1",2,2,"sm_120a","+ptx87",null,[4593,null,null,4594,null],null,0]
+["mma_m16n8k32_f16_e4m3_e2m3",2,1,null,null,null,[4595,4596,4597,null,null],null,0]
+["mma_m16n8k32_f16_e4m3_e2m3",2,2,"sm_120a","+ptx87",null,[4598,null,null,4599,null],null,0]
+["mma_m16n8k32_f16_e4m3_e3m2",2,1,null,null,null,[4600,4601,4602,null,null],null,0]
+["mma_m16n8k32_f16_e4m3_e3m2",2,2,"sm_120a","+ptx87",null,[4603,null,null,4604,null],null,0]
+["mma_m16n8k32_f16_e4m3_e4m3",2,1,null,null,null,[4605,4606,4607,null,null],null,0]
+["mma_m16n8k32_f16_e4m3_e4m3",2,2,"sm_120a","+ptx87",null,[4608,null,null,4609,null],null,0]
+["mma_m16n8k32_f16_e4m3_e5m2",2,1,null,null,null,[4610,4611,4612,null,null],null,0]
+["mma_m16n8k32_f16_e4m3_e5m2",2,2,"sm_120a","+ptx87",null,[4613,null,null,4614,null],null,0]
+["mma_m16n8k32_f16_e5m2_e2m1",2,1,null,null,null,[4615,4616,4617,null,null],null,0]
+["mma_m16n8k32_f16_e5m2_e2m1",2,2,"sm_120a","+ptx87",null,[4618,null,null,4619,null],null,0]
+["mma_m16n8k32_f16_e5m2_e2m3",2,1,null,null,null,[4620,4621,4622,null,null],null,0]
+["mma_m16n8k32_f16_e5m2_e2m3",2,2,"sm_120a","+ptx87",null,[4623,null,null,4624,null],null,0]
+["mma_m16n8k32_f16_e5m2_e3m2",2,1,null,null,null,[4625,4626,4627,null,null],null,0]
+["mma_m16n8k32_f16_e5m2_e3m2",2,2,"sm_120a","+ptx87",null,[4628,null,null,4629,null],null,0]
+["mma_m16n8k32_f16_e5m2_e4m3",2,1,null,null,null,[4630,4631,4632,null,null],null,0]
+["mma_m16n8k32_f16_e5m2_e4m3",2,2,"sm_120a","+ptx87",null,[4633,null,null,4634,null],null,0]
+["mma_m16n8k32_f16_e5m2_e5m2",2,1,null,null,null,[4635,4636,4637,null,null],null,0]
+["mma_m16n8k32_f16_e5m2_e5m2",2,2,"sm_120a","+ptx87",null,[4638,null,null,4639,null],null,0]
+["mma_m16n8k32_f32_e2m1_e2m1",2,1,null,null,null,[4640,4641,4642,null,null],null,0]
+["mma_m16n8k32_f32_e2m1_e2m1",2,2,"sm_120a","+ptx87",null,[4643,null,null,4644,null],null,0]
+["mma_m16n8k32_f32_e2m1_e2m3",2,1,null,null,null,[4645,4646,4647,null,null],null,0]
+["mma_m16n8k32_f32_e2m1_e2m3",2,2,"sm_120a","+ptx87",null,[4648,null,null,4649,null],null,0]
+["mma_m16n8k32_f32_e2m1_e3m2",2,1,null,null,null,[4650,4651,4652,null,null],null,0]
+["mma_m16n8k32_f32_e2m1_e3m2",2,2,"sm_120a","+ptx87",null,[4653,null,null,4654,null],null,0]
+["mma_m16n8k32_f32_e2m1_e4m3",2,1,null,null,null,[4655,4656,4657,null,null],null,0]
+["mma_m16n8k32_f32_e2m1_e4m3",2,2,"sm_120a","+ptx87",null,[4658,null,null,4659,null],null,0]
+["mma_m16n8k32_f32_e2m1_e5m2",2,1,null,null,null,[4660,4661,4662,null,null],null,0]
+["mma_m16n8k32_f32_e2m1_e5m2",2,2,"sm_120a","+ptx87",null,[4663,null,null,4664,null],null,0]
+["mma_m16n8k32_f32_e2m3_e2m1",2,1,null,null,null,[4665,4666,4667,null,null],null,0]
+["mma_m16n8k32_f32_e2m3_e2m1",2,2,"sm_120a","+ptx87",null,[4668,null,null,4669,null],null,0]
+["mma_m16n8k32_f32_e2m3_e2m3",2,1,null,null,null,[4670,4671,4672,null,null],null,0]
+["mma_m16n8k32_f32_e2m3_e2m3",2,2,"sm_120a","+ptx87",null,[4673,null,null,4674,null],null,0]
+["mma_m16n8k32_f32_e2m3_e3m2",2,1,null,null,null,[4675,4676,4677,null,null],null,0]
+["mma_m16n8k32_f32_e2m3_e3m2",2,2,"sm_120a","+ptx87",null,[4678,null,null,4679,null],null,0]
+["mma_m16n8k32_f32_e2m3_e4m3",2,1,null,null,null,[4680,4681,4682,null,null],null,0]
+["mma_m16n8k32_f32_e2m3_e4m3",2,2,"sm_120a","+ptx87",null,[4683,null,null,4684,null],null,0]
+["mma_m16n8k32_f32_e2m3_e5m2",2,1,null,null,null,[4685,4686,4687,null,null],null,0]
+["mma_m16n8k32_f32_e2m3_e5m2",2,2,"sm_120a","+ptx87",null,[4688,null,null,4689,null],null,0]
+["mma_m16n8k32_f32_e3m2_e2m1",2,1,null,null,null,[4690,4691,4692,null,null],null,0]
+["mma_m16n8k32_f32_e3m2_e2m1",2,2,"sm_120a","+ptx87",null,[4693,null,null,4694,null],null,0]
+["mma_m16n8k32_f32_e3m2_e2m3",2,1,null,null,null,[4695,4696,4697,null,null],null,0]
+["mma_m16n8k32_f32_e3m2_e2m3",2,2,"sm_120a","+ptx87",null,[4698,null,null,4699,null],null,0]
+["mma_m16n8k32_f32_e3m2_e3m2",2,1,null,null,null,[4700,4701,4702,null,null],null,0]
+["mma_m16n8k32_f32_e3m2_e3m2",2,2,"sm_120a","+ptx87",null,[4703,null,null,4704,null],null,0]
+["mma_m16n8k32_f32_e3m2_e4m3",2,1,null,null,null,[4705,4706,4707,null,null],null,0]
+["mma_m16n8k32_f32_e3m2_e4m3",2,2,"sm_120a","+ptx87",null,[4708,null,null,4709,null],null,0]
+["mma_m16n8k32_f32_e3m2_e5m2",2,1,null,null,null,[4710,4711,4712,null,null],null,0]
+["mma_m16n8k32_f32_e3m2_e5m2",2,2,"sm_120a","+ptx87",null,[4713,null,null,4714,null],null,0]
+["mma_m16n8k32_f32_e4m3_e2m1",2,1,null,null,null,[4715,4716,4717,null,null],null,0]
+["mma_m16n8k32_f32_e4m3_e2m1",2,2,"sm_120a","+ptx87",null,[4718,null,null,4719,null],null,0]
+["mma_m16n8k32_f32_e4m3_e2m3",2,1,null,null,null,[4720,4721,4722,null,null],null,0]
+["mma_m16n8k32_f32_e4m3_e2m3",2,2,"sm_120a","+ptx87",null,[4723,null,null,4724,null],null,0]
+["mma_m16n8k32_f32_e4m3_e3m2",2,1,null,null,null,[4725,4726,4727,null,null],null,0]
+["mma_m16n8k32_f32_e4m3_e3m2",2,2,"sm_120a","+ptx87",null,[4728,null,null,4729,null],null,0]
+["mma_m16n8k32_f32_e4m3_e4m3",2,1,null,null,null,[4730,4731,4732,null,null],null,0]
+["mma_m16n8k32_f32_e4m3_e4m3",2,2,"sm_120a","+ptx87",null,[4733,null,null,4734,null],null,0]
+["mma_m16n8k32_f32_e4m3_e5m2",2,1,null,null,null,[4735,4736,4737,null,null],null,0]
+["mma_m16n8k32_f32_e4m3_e5m2",2,2,"sm_120a","+ptx87",null,[4738,null,null,4739,null],null,0]
+["mma_m16n8k32_f32_e5m2_e2m1",2,1,null,null,null,[4740,4741,4742,null,null],null,0]
+["mma_m16n8k32_f32_e5m2_e2m1",2,2,"sm_120a","+ptx87",null,[4743,null,null,4744,null],null,0]
+["mma_m16n8k32_f32_e5m2_e2m3",2,1,null,null,null,[4745,4746,4747,null,null],null,0]
+["mma_m16n8k32_f32_e5m2_e2m3",2,2,"sm_120a","+ptx87",null,[4748,null,null,4749,null],null,0]
+["mma_m16n8k32_f32_e5m2_e3m2",2,1,null,null,null,[4750,4751,4752,null,null],null,0]
+["mma_m16n8k32_f32_e5m2_e3m2",2,2,"sm_120a","+ptx87",null,[4753,null,null,4754,null],null,0]
+["mma_m16n8k32_f32_e5m2_e4m3",2,1,null,null,null,[4755,4756,4757,null,null],null,0]
+["mma_m16n8k32_f32_e5m2_e4m3",2,2,"sm_120a","+ptx87",null,[4758,null,null,4759,null],null,0]
+["mma_m16n8k32_f32_e5m2_e5m2",2,1,null,null,null,[4760,4761,4762,null,null],null,0]
+["mma_m16n8k32_f32_e5m2_e5m2",2,2,"sm_120a","+ptx87",null,[4763,null,null,4764,null],null,0]
+["mma_m16n8k32_fp8_f16_e4m3_e4m3",2,1,null,null,null,[4765,4766,4767,null,null],null,0]
+["mma_m16n8k32_fp8_f16_e4m3_e5m2",2,1,null,null,null,[4768,4769,4770,null,null],null,0]
+["mma_m16n8k32_fp8_f16_e5m2_e4m3",2,1,null,null,null,[4771,4772,4773,null,null],null,0]
+["mma_m16n8k32_fp8_f16_e5m2_e5m2",2,1,null,null,null,[4774,4775,4776,null,null],null,0]
+["mma_m16n8k32_fp8_f32_e4m3_e4m3",2,1,null,null,null,[4777,4778,4779,null,null],null,0]
+["mma_m16n8k32_fp8_f32_e4m3_e5m2",2,1,null,null,null,[4780,4781,4782,null,null],null,0]
+["mma_m16n8k32_fp8_f32_e5m2_e4m3",2,1,null,null,null,[4783,4784,4785,null,null],null,0]
+["mma_m16n8k32_fp8_f32_e5m2_e5m2",2,1,null,null,null,[4786,4787,4788,null,null],null,0]
+["mma_m16n8k4_f32_tf32",2,1,null,null,null,[4789,4790,4791,null,null],null,0]
+["mma_m16n8k8_f16_f16",2,1,null,null,null,[4792,4793,4794,null,null],null,0]
+["mma_m16n8k8_f32_bf16",2,1,null,null,null,[4795,4796,4797,null,null],null,0]
+["mma_m16n8k8_f32_f16",2,1,null,null,null,[4798,4799,4800,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16",2,1,null,null,null,[4801,4802,4803,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16",2,1,null,null,null,[4804,4805,4806,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16",2,1,null,null,null,[4807,4808,4809,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16",2,1,null,null,null,[4810,4811,4812,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16",2,1,null,null,null,[4813,4814,4815,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16",2,1,null,null,null,[4816,4817,4818,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16",2,1,null,null,null,[4819,4820,4821,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16",2,1,null,null,null,[4822,4823,4824,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16",2,1,null,null,null,[4825,4826,4827,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16",2,1,null,null,null,[4828,4829,4830,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16",2,1,null,null,null,[4831,4832,4833,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16",2,1,null,null,null,[4834,4835,4836,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16",2,1,null,null,null,[4837,4838,4839,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16",2,1,null,null,null,[4840,4841,4842,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16",2,1,null,null,null,[4843,4844,4845,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16",2,1,null,null,null,[4846,4847,4848,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16",2,1,null,null,null,[4849,4850,4851,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16",2,1,null,null,null,[4852,4853,4854,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16",2,1,null,null,null,[4855,4856,4857,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16",2,1,null,null,null,[4858,4859,4860,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16",2,1,null,null,null,[4861,4862,4863,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16",2,1,null,null,null,[4864,4865,4866,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16",2,1,null,null,null,[4867,4868,4869,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16",2,1,null,null,null,[4870,4871,4872,null,null],null,0]
+["mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16",2,1,null,null,null,[4873,4874,4875,null,null],null,0]
+["rsqrt_approx_f32",2,1,null,null,null,[4876,4877,4878,null,null],null,0]
+["rsqrt_approx_f64",2,1,null,null,null,[4879,4880,4881,null,null],null,0]
+["rsqrt_approx_ftz_f32",2,1,null,null,null,[4882,4883,4884,null,null],null,0]
+["sin_approx_f32",2,1,null,null,null,[4885,4886,4887,null,null],null,0]
+["sin_approx_ftz_f32",2,1,null,null,null,[4888,4889,4890,null,null],null,0]
+["cvt_rn_satfinite_e4m3x2_f32",3,2,"sm_89","+ptx81",null,[4891,null,null,4892,null],null,0]
+["cvt_rn_satfinite_e4m3x2_f32",3,3,"sm_89","+ptx81","sm_89",[4891,null,null,4892,714],null,0]
+["cvt_rn_satfinite_e5m2x2_f32",3,2,"sm_89","+ptx81",null,[4893,null,null,4894,null],null,0]
+["cvt_rn_satfinite_e5m2x2_f32",3,3,"sm_89","+ptx81","sm_89",[4893,null,null,4894,714],null,0]
+["cvt_rn_satfinite_relu_e4m3x2_f32",3,2,"sm_89","+ptx81",null,[4895,null,null,4896,null],null,0]
+["cvt_rn_satfinite_relu_e4m3x2_f32",3,3,"sm_89","+ptx81","sm_89",[4895,null,null,4896,714],null,0]
+["cvt_rn_satfinite_relu_e5m2x2_f32",3,2,"sm_89","+ptx81",null,[4897,null,null,4898,null],null,0]
+["cvt_rn_satfinite_relu_e5m2x2_f32",3,3,"sm_89","+ptx81","sm_89",[4897,null,null,4898,714],null,0]
diff --git a/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/inputs.jsonl b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/inputs.jsonl
new file mode 100644
index 0000000000..481ccb97cd
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/inputs.jsonl
@@ -0,0 +1,307 @@
+[4098,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rm.ftz.sat.f(float, float)\n"]
+[4101,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rm.sat.f(float, float)\n"]
+[4104,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rn.ftz.sat.f(float, float)\n"]
+[4107,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rn.sat.f(float, float)\n"]
+[4110,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rp.ftz.sat.f(float, float)\n"]
+[4113,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rp.sat.f(float, float)\n"]
+[4116,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rz.ftz.sat.f(float, float)\n"]
+[4119,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.add.rz.sat.f(float, float)\n"]
+[4122,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.cos.approx.f(float)\n"]
+[4125,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.cos.approx.ftz.f(float)\n"]
+[4128,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.ff2f16x2.rn(float, float)\n"]
+[4131,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.ff2f16x2.rn(float, float)\n\ndefine <2 x half> @auxiliary_cvt_f16x2_f32(float %arg0, float %arg1) {\n  %result = call <2 x half> @llvm.nvvm.ff2f16x2.rn(float %arg0, float %arg1)\n  ret <2 x half> %result\n}\n"]
+[4133,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.ff2bf16x2.rn(float, float)\n"]
+[4136,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.ff2bf16x2.rn(float, float)\n\ndefine <2 x bfloat> @auxiliary_cvt_f32x2_bf16x2(float %arg0, float %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.ff2bf16x2.rn(float %arg0, float %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4138,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn(i16)\n"]
+[4141,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn(i16)\n\ndefine <2 x half> @auxiliary_cvt_rn_f16x2_e4m3x2(i16 %arg0) {\n  %result = call <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn(i16 %arg0)\n  ret <2 x half> %result\n}\n"]
+[4143,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn(i16)\n"]
+[4146,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn(i16)\n\ndefine <2 x half> @auxiliary_cvt_rn_f16x2_e5m2x2(i16 %arg0) {\n  %result = call <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn(i16 %arg0)\n  ret <2 x half> %result\n}\n"]
+[4148,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.ff2bf16x2.rn.relu(float, float)\n"]
+[4151,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.ff2bf16x2.rn.relu(float, float)\n\ndefine <2 x bfloat> @auxiliary_cvt_rn_relu_bf16x2_f32(float %arg0, float %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.ff2bf16x2.rn.relu(float %arg0, float %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4153,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn.relu(i16)\n"]
+[4156,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn.relu(i16)\n\ndefine <2 x half> @auxiliary_cvt_rn_relu_f16x2_e4m3x2(i16 %arg0) {\n  %result = call <2 x half> @llvm.nvvm.e4m3x2.to.f16x2.rn.relu(i16 %arg0)\n  ret <2 x half> %result\n}\n"]
+[4158,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn.relu(i16)\n"]
+[4161,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn.relu(i16)\n\ndefine <2 x half> @auxiliary_cvt_rn_relu_f16x2_e5m2x2(i16 %arg0) {\n  %result = call <2 x half> @llvm.nvvm.e5m2x2.to.f16x2.rn.relu(i16 %arg0)\n  ret <2 x half> %result\n}\n"]
+[4163,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.ff2f16x2.rn.relu(float, float)\n"]
+[4166,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.ff2f16x2.rn.relu(float, float)\n\ndefine <2 x half> @auxiliary_cvt_rn_relu_f16x2_f32(float %arg0, float %arg1) {\n  %result = call <2 x half> @llvm.nvvm.ff2f16x2.rn.relu(float %arg0, float %arg1)\n  ret <2 x half> %result\n}\n"]
+[4168,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e4m3x2.rn(<2 x half>)\n"]
+[4171,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e4m3x2.rn(<2 x half>)\n\ndefine i16 @auxiliary_cvt_rn_satfinite_e4m3x2_f16x2(<2 x half> %arg0) {\n  %result = call i16 @llvm.nvvm.f16x2.to.e4m3x2.rn(<2 x half> %arg0)\n  ret i16 %result\n}\n"]
+[4173,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e5m2x2.rn(<2 x half>)\n"]
+[4176,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e5m2x2.rn(<2 x half>)\n\ndefine i16 @auxiliary_cvt_rn_satfinite_e5m2x2_f16x2(<2 x half> %arg0) {\n  %result = call i16 @llvm.nvvm.f16x2.to.e5m2x2.rn(<2 x half> %arg0)\n  ret i16 %result\n}\n"]
+[4178,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e4m3x2.rn.relu(<2 x half>)\n"]
+[4181,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e4m3x2.rn.relu(<2 x half>)\n\ndefine i16 @auxiliary_cvt_rn_satfinite_relu_e4m3x2_f16x2(<2 x half> %arg0) {\n  %result = call i16 @llvm.nvvm.f16x2.to.e4m3x2.rn.relu(<2 x half> %arg0)\n  ret i16 %result\n}\n"]
+[4183,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e5m2x2.rn.relu(<2 x half>)\n"]
+[4186,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare i16 @llvm.nvvm.f16x2.to.e5m2x2.rn.relu(<2 x half>)\n\ndefine i16 @auxiliary_cvt_rn_satfinite_relu_e5m2x2_f16x2(<2 x half> %arg0) {\n  %result = call i16 @llvm.nvvm.f16x2.to.e5m2x2.rn.relu(<2 x half> %arg0)\n  ret i16 %result\n}\n"]
+[4188,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.ff2bf16x2.rz(float, float)\n"]
+[4191,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.ff2bf16x2.rz(float, float)\n\ndefine <2 x bfloat> @auxiliary_cvt_rz_bf16x2_f32(float %arg0, float %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.ff2bf16x2.rz(float %arg0, float %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4193,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.ff2f16x2.rz(float, float)\n"]
+[4196,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.ff2f16x2.rz(float, float)\n\ndefine <2 x half> @auxiliary_cvt_rz_f16x2_f32(float %arg0, float %arg1) {\n  %result = call <2 x half> @llvm.nvvm.ff2f16x2.rz(float %arg0, float %arg1)\n  ret <2 x half> %result\n}\n"]
+[4198,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rm.ftz.sat.f(float, float, float)\n"]
+[4201,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rm.sat.f(float, float, float)\n"]
+[4204,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rn.ftz.sat.f(float, float, float)\n"]
+[4207,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rn.sat.f(float, float, float)\n"]
+[4210,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rp.ftz.sat.f(float, float, float)\n"]
+[4213,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rp.sat.f(float, float, float)\n"]
+[4216,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rz.ftz.sat.f(float, float, float)\n"]
+[4219,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fma.rz.sat.f(float, float, float)\n"]
+[4222,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.lg2.approx.f(float)\n"]
+[4225,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.lg2.approx.ftz.f(float)\n"]
+[4228,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.bf16(bfloat, bfloat)\n"]
+[4231,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_max_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmax.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4233,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.f16(half, half)\n"]
+[4236,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.f16(half, half)\n\ndefine half @auxiliary_max_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4238,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.f16(half, half)\n"]
+[4241,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.f16(half, half)\n\ndefine half @auxiliary_max_ftz_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.ftz.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4243,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.f16x2(<2 x half>, <2 x half>)\n"]
+[4246,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_ftz_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.ftz.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4248,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.nan.f16(half, half)\n"]
+[4251,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.nan.f16(half, half)\n\ndefine half @auxiliary_max_ftz_nan_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.ftz.nan.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4253,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.nan.f16x2(<2 x half>, <2 x half>)\n"]
+[4256,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.nan.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_ftz_nan_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.ftz.nan.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4258,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16(half, half)\n"]
+[4261,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_max_ftz_nan_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4263,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4266,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_ftz_nan_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4268,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f(float, float)\n"]
+[4271,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_max_ftz_nan_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmax.ftz.nan.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4273,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.xorsign.abs.f16(half, half)\n"]
+[4276,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.ftz.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_max_ftz_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.ftz.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4278,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4281,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.ftz.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_ftz_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.ftz.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4283,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.ftz.xorsign.abs.f(float, float)\n"]
+[4286,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.ftz.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_max_ftz_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmax.ftz.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4288,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.nan.bf16(bfloat, bfloat)\n"]
+[4291,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.nan.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_max_nan_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmax.nan.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4293,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmax.nan.bf16x2(<2 x bfloat>, <2 x bfloat>)\n"]
+[4296,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmax.nan.bf16x2(<2 x bfloat>, <2 x bfloat>)\n\ndefine <2 x bfloat> @auxiliary_max_nan_bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.fmax.nan.bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4298,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.nan.f16(half, half)\n"]
+[4301,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.nan.f16(half, half)\n\ndefine half @auxiliary_max_nan_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.nan.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4303,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.nan.f16x2(<2 x half>, <2 x half>)\n"]
+[4306,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.nan.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_nan_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.nan.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4308,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.nan.xorsign.abs.bf16(bfloat, bfloat)\n"]
+[4311,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.nan.xorsign.abs.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_max_nan_xorsign_abs_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmax.nan.xorsign.abs.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4313,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmax.nan.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n"]
+[4316,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmax.nan.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n\ndefine <2 x bfloat> @auxiliary_max_nan_xorsign_abs_bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.fmax.nan.xorsign.abs.bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4318,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.nan.xorsign.abs.f16(half, half)\n"]
+[4321,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.nan.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_max_nan_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.nan.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4323,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4326,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_nan_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.nan.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4328,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.nan.xorsign.abs.f(float, float)\n"]
+[4331,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.nan.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_max_nan_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmax.nan.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4333,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.xorsign.abs.bf16(bfloat, bfloat)\n"]
+[4336,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmax.xorsign.abs.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_max_xorsign_abs_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmax.xorsign.abs.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4338,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmax.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n"]
+[4341,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmax.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n\ndefine <2 x bfloat> @auxiliary_max_xorsign_abs_bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.fmax.xorsign.abs.bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4343,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.xorsign.abs.f16(half, half)\n"]
+[4346,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmax.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_max_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmax.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4348,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4351,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmax.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_max_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmax.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4353,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.xorsign.abs.f(float, float)\n"]
+[4356,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmax.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_max_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmax.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4358,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.bf16(bfloat, bfloat)\n"]
+[4361,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_min_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmin.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4363,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.f16(half, half)\n"]
+[4366,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.f16(half, half)\n\ndefine half @auxiliary_min_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4368,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.f16(half, half)\n"]
+[4371,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.f16(half, half)\n\ndefine half @auxiliary_min_ftz_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.ftz.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4373,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.f16x2(<2 x half>, <2 x half>)\n"]
+[4376,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_ftz_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.ftz.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4378,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.nan.f16(half, half)\n"]
+[4381,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.nan.f16(half, half)\n\ndefine half @auxiliary_min_ftz_nan_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.ftz.nan.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4383,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.nan.f16x2(<2 x half>, <2 x half>)\n"]
+[4386,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.nan.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_ftz_nan_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.ftz.nan.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4388,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16(half, half)\n"]
+[4391,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_min_ftz_nan_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4393,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4396,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_ftz_nan_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4398,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f(float, float)\n"]
+[4401,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_min_ftz_nan_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmin.ftz.nan.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4403,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.xorsign.abs.f16(half, half)\n"]
+[4406,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.ftz.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_min_ftz_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.ftz.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4408,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4411,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.ftz.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_ftz_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.ftz.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4413,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.ftz.xorsign.abs.f(float, float)\n"]
+[4416,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.ftz.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_min_ftz_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmin.ftz.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4418,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.nan.bf16(bfloat, bfloat)\n"]
+[4421,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.nan.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_min_nan_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmin.nan.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4423,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmin.nan.bf16x2(<2 x bfloat>, <2 x bfloat>)\n"]
+[4426,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmin.nan.bf16x2(<2 x bfloat>, <2 x bfloat>)\n\ndefine <2 x bfloat> @auxiliary_min_nan_bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.fmin.nan.bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4428,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.nan.f16(half, half)\n"]
+[4431,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.nan.f16(half, half)\n\ndefine half @auxiliary_min_nan_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.nan.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4433,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.nan.f16x2(<2 x half>, <2 x half>)\n"]
+[4436,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.nan.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_nan_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.nan.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4438,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.nan.xorsign.abs.bf16(bfloat, bfloat)\n"]
+[4441,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.nan.xorsign.abs.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_min_nan_xorsign_abs_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmin.nan.xorsign.abs.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4443,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmin.nan.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n"]
+[4446,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmin.nan.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n\ndefine <2 x bfloat> @auxiliary_min_nan_xorsign_abs_bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.fmin.nan.xorsign.abs.bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4448,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.nan.xorsign.abs.f16(half, half)\n"]
+[4451,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.nan.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_min_nan_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.nan.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4453,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4456,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.nan.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_nan_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.nan.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4458,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.nan.xorsign.abs.f(float, float)\n"]
+[4461,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.nan.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_min_nan_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmin.nan.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4463,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.xorsign.abs.bf16(bfloat, bfloat)\n"]
+[4466,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare bfloat @llvm.nvvm.fmin.xorsign.abs.bf16(bfloat, bfloat)\n\ndefine bfloat @auxiliary_min_xorsign_abs_bf16(bfloat %arg0, bfloat %arg1) {\n  %result = call bfloat @llvm.nvvm.fmin.xorsign.abs.bf16(bfloat %arg0, bfloat %arg1)\n  ret bfloat %result\n}\n"]
+[4468,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmin.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n"]
+[4471,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x bfloat> @llvm.nvvm.fmin.xorsign.abs.bf16x2(<2 x bfloat>, <2 x bfloat>)\n\ndefine <2 x bfloat> @auxiliary_min_xorsign_abs_bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1) {\n  %result = call <2 x bfloat> @llvm.nvvm.fmin.xorsign.abs.bf16x2(<2 x bfloat> %arg0, <2 x bfloat> %arg1)\n  ret <2 x bfloat> %result\n}\n"]
+[4473,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.xorsign.abs.f16(half, half)\n"]
+[4476,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare half @llvm.nvvm.fmin.xorsign.abs.f16(half, half)\n\ndefine half @auxiliary_min_xorsign_abs_f16(half %arg0, half %arg1) {\n  %result = call half @llvm.nvvm.fmin.xorsign.abs.f16(half %arg0, half %arg1)\n  ret half %result\n}\n"]
+[4478,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n"]
+[4481,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare <2 x half> @llvm.nvvm.fmin.xorsign.abs.f16x2(<2 x half>, <2 x half>)\n\ndefine <2 x half> @auxiliary_min_xorsign_abs_f16x2(<2 x half> %arg0, <2 x half> %arg1) {\n  %result = call <2 x half> @llvm.nvvm.fmin.xorsign.abs.f16x2(<2 x half> %arg0, <2 x half> %arg1)\n  ret <2 x half> %result\n}\n"]
+[4483,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.xorsign.abs.f(float, float)\n"]
+[4486,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.fmin.xorsign.abs.f(float, float)\n\ndefine float @auxiliary_min_xorsign_abs_f32(float %arg0, float %arg1) {\n  %result = call float @llvm.nvvm.fmin.xorsign.abs.f(float %arg0, float %arg1)\n  ret float %result\n}\n"]
+[4488,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k16.row.col.f16.f16(<2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>)\n"]
+[4491,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k16.row.col.f16.e4m3.e4m3.f16(i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4494,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k16.row.col.f16.e4m3.e5m2.f16(i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4497,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k16.row.col.f16.e5m2.e4m3.f16(i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4500,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k16.row.col.f16.e5m2.e5m2.f16(i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4503,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k16.row.col.f32.e4m3.e4m3.f32(i32, i32, i32, float, float, float, float)\n"]
+[4506,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k16.row.col.f32.e4m3.e5m2.f32(i32, i32, i32, float, float, float, float)\n"]
+[4509,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k16.row.col.f32.e5m2.e4m3.f32(i32, i32, i32, float, float, float, float)\n"]
+[4512,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k16.row.col.f32.e5m2.e5m2.f32(i32, i32, i32, float, float, float, float)\n"]
+[4515,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4518,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m1_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4520,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4523,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m1_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4525,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4528,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m1_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4530,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4533,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m1_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4535,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4538,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m1_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4540,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4543,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m3_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4545,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4548,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m3_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4550,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4553,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m3_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4555,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4558,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m3_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4560,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4563,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e2m3_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4565,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4568,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e3m2_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4570,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4573,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e3m2_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4575,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4578,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e3m2_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4580,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4583,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e3m2_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4585,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4588,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e3m2_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4590,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4593,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e4m3_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4595,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4598,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e4m3_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4600,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4603,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e4m3_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4605,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4608,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e4m3_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4610,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4613,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e4m3_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4615,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4618,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e5m2_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4620,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4623,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e5m2_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4625,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4628,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e5m2_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4630,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4633,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e5m2_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4635,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4638,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n\ndefine { <2 x half>, <2 x half> } @auxiliary_mma_m16n8k32_f16_e5m2_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7) {\n  %result = call { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, <2 x half> %arg6, <2 x half> %arg7)\n  ret { <2 x half>, <2 x half> } %result\n}\n"]
+[4640,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4643,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m1_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4645,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4648,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m1_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4650,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4653,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m1_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4655,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4658,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m1_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4660,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4663,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m1_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4665,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4668,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m3_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4670,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4673,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m3_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4675,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4678,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m3_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4680,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4683,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m3_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4685,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4688,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e2m3_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4690,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4693,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e3m2_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4695,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4698,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e3m2_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4700,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4703,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e3m2_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4705,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4708,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e3m2_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4710,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4713,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e3m2_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4715,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4718,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e4m3_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4720,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4723,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e4m3_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4725,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4728,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e4m3_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4730,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4733,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e4m3_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4735,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4738,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e4m3_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4740,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4743,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e5m2_e2m1(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4745,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4748,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e5m2_e2m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4750,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4753,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e5m2_e3m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4755,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4758,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e5m2_e4m3(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4760,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4763,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n\ndefine { float, float, float, float } @auxiliary_mma_m16n8k32_f32_e5m2_e5m2(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9) {\n  %result = call { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32(i32 %arg0, i32 %arg1, i32 %arg2, i32 %arg3, i32 %arg4, i32 %arg5, float %arg6, float %arg7, float %arg8, float %arg9)\n  ret { float, float, float, float } %result\n}\n"]
+[4765,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.f16.e4m3.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4768,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.f16.e4m3.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4771,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.f16.e5m2.e4m3.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4774,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k32.row.col.f16.e5m2.e5m2.f16(i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>)\n"]
+[4777,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.f32.e4m3.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4780,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.f32.e4m3.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4783,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.f32.e5m2.e4m3.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4786,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k32.row.col.f32.e5m2.e5m2.f32(i32, i32, i32, i32, i32, i32, float, float, float, float)\n"]
+[4789,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k4.row.col.tf32(i32, i32, i32, float, float, float, float)\n"]
+[4792,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.m16n8k8.row.col.f16.f16(<2 x half>, <2 x half>, <2 x half>, <2 x half>, <2 x half>)\n"]
+[4795,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k8.row.col.bf16(i32, i32, i32, float, float, float, float)\n"]
+[4798,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { float, float, float, float } @llvm.nvvm.mma.m16n8k8.row.col.f32.f32(<2 x half>, <2 x half>, <2 x half>, float, float, float, float)\n"]
+[4801,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4804,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4807,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4810,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4813,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4816,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4819,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4822,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4825,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4828,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4831,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4834,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4837,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4840,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4843,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4846,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4849,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4852,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4855,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4858,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4861,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4864,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4867,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4870,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4873,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare { <2 x half>, <2 x half> } @llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16(i32, i32, i32, i32, i32, i32, i32, i32, <2 x half>, <2 x half>, i32, i32)\n"]
+[4876,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.rsqrt.approx.f(float)\n"]
+[4879,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare double @llvm.nvvm.rsqrt.approx.d(double)\n"]
+[4882,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.rsqrt.approx.ftz.f(float)\n"]
+[4885,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.sin.approx.f(float)\n"]
+[4888,"; Auxiliary typed-NVVM validation fixture; dynamic arguments retain the intrinsic.\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndeclare float @llvm.nvvm.sin.approx.ftz.f(float)\n"]
+[4891,"; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n; SPDX-License-Identifier: Apache-2.0\n; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.\n; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: bd6f78b594b948325a3c193b8d99df05ae1fe296bc2257ce67abd9932a59d1f9.\n; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.\n\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndefine i16 @probe_cvt_rn_satfinite_e4m3x2_f32(float %low, float %high) {\n  %result = call i16 asm \"cvt.rn.satfinite.e4m3x2.f32 $0, $2, $1;\", \"=h,f,f\"(float %low, float %high)\n  ret i16 %result\n}\n"]
+[4893,"; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n; SPDX-License-Identifier: Apache-2.0\n; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.\n; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: bd6f78b594b948325a3c193b8d99df05ae1fe296bc2257ce67abd9932a59d1f9.\n; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.\n\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndefine i16 @probe_cvt_rn_satfinite_e5m2x2_f32(float %low, float %high) {\n  %result = call i16 asm \"cvt.rn.satfinite.e5m2x2.f32 $0, $2, $1;\", \"=h,f,f\"(float %low, float %high)\n  ret i16 %result\n}\n"]
+[4895,"; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n; SPDX-License-Identifier: Apache-2.0\n; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.\n; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: bd6f78b594b948325a3c193b8d99df05ae1fe296bc2257ce67abd9932a59d1f9.\n; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.\n\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndefine i16 @probe_cvt_rn_satfinite_relu_e4m3x2_f32(float %low, float %high) {\n  %result = call i16 asm \"cvt.rn.satfinite.relu.e4m3x2.f32 $0, $2, $1;\", \"=h,f,f\"(float %low, float %high)\n  ret i16 %result\n}\n"]
+[4897,"; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\n; SPDX-License-Identifier: Apache-2.0\n; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.\n; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: bd6f78b594b948325a3c193b8d99df05ae1fe296bc2257ce67abd9932a59d1f9.\n; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.\n\ntarget triple = \"nvptx64-nvidia-cuda\"\n\ndefine i16 @probe_cvt_rn_satfinite_relu_e5m2x2_f32(float %low, float %high) {\n  %result = call i16 asm \"cvt.rn.satfinite.relu.e5m2x2.f32 $0, $2, $1;\", \"=h,f,f\"(float %low, float %high)\n  ret i16 %result\n}\n"]
diff --git a/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/metadata.json b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/metadata.json
new file mode 100644
index 0000000000..9e14c8f494
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/metadata.json
@@ -0,0 +1,411 @@
+{
+  "schema": 1,
+  "date": "2026-10-11",
+  "summary": {
+    "primary_captures": 1699,
+    "primary_intrinsic_ids": 1029,
+    "primary_terminal_captures": 1506,
+    "initial_candidate_failures": 192,
+    "initial_renderer_panics": 148,
+    "initial_invalid_target_requests": 44,
+    "primary_split_target_captures": 56,
+    "auxiliary_captures": 377,
+    "auxiliary_typed_declarations": 187,
+    "auxiliary_typed_codegen": 116,
+    "auxiliary_typed_assemblies": 66,
+    "auxiliary_inline_codegen": 4,
+    "auxiliary_inline_assemblies": 4,
+    "stage_groups": 770,
+    "stage_artifact_proofs": 3753,
+    "historical_stage_groups_not_revalidated": 38,
+    "failures": 0,
+    "unique_artifacts": 4899,
+    "persisted_auxiliary_inputs": 307
+  },
+  "tools": {
+    "llvm-as": {
+      "path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llvm-as",
+      "sha256": "6a7d24f098fe0df8a735e269cbf407a2322ae04b84cb4898c77367da898a1cb3",
+      "version": "LLVM version 23.1.3-rust-1.101.0-nightly"
+    },
+    "llvm-dis": {
+      "path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llvm-dis",
+      "sha256": "e2dfff788815d3f75709cf4fbbf237b98764b52baa191159f9100888197e934e",
+      "version": "LLVM version 23.1.3-rust-1.101.0-nightly"
+    },
+    "llc": {
+      "path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+      "sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+      "version": "LLVM version 23.1.3-rust-1.101.0-nightly"
+    },
+    "ptxas": {
+      "path": "/usr/local/cuda-13.3/bin/ptxas",
+      "sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc",
+      "version": "Cuda compilation tools, release 13.3, V13.3.73",
+      "build": "cuda_13.3.r13.3/compiler.38244171_0"
+    },
+    "llvm-tblgen": {
+      "path": "/tmp/cudaoxide-intrinsics-tblgen-build-20261011/bin/llvm-tblgen",
+      "version": "LLVM version 23.1.3",
+      "sha256": "f9e28cd046b8084efed066634a239076763ac7a7a6bde932b58e2cb439c25564"
+    },
+    "llvm-tblgen-22": {
+      "name": "llvm-tblgen-22",
+      "version_line": "Ubuntu LLVM version 22.1.8",
+      "sha256": "dbe41ce85b55b137b846da027f34bdeb260d25b907059a39927eea6aee4cae96",
+      "enforce_sha256": false,
+      "provenance": "Ubuntu package binary used only for a byte-for-byte dump comparison; it reproduced the Intrinsics.td dump hash exactly. The NVPTX.td dump uses LLVM 23-only TableGen operators (!sort in Target.td) that pre-23 binaries reject, so no independent binary was available for that dump"
+    }
+  },
+  "source": {
+    "llvm_repository": "https://github.com/rust-lang/llvm-project.git",
+    "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+    "llvm_git_tree": "743426f16c6d14d89426972b8aad1540e5d25ab0",
+    "rust_commit": "6f5da406b56c2db6607202b63daee73aa8c61297",
+    "dump_hashes": {
+      "intrinsics_sha256": "c0fb05eda32f2fe89ba600d4d2dc882e0bf54d5aa60b5e5f8601eb87a9881408",
+      "nvptx_sha256": "492011cd3c20fbd888a2fa05e7eb5d894c0db79270c5bf6dbda0b6648d3c6582",
+      "normalized_imported_sha256": "54d47b8e281dd953dc95fe16f834ff38164d9809171c9a1af097231b0bc45cab"
+    },
+    "normalized_declarations": 2633,
+    "old_llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
+    "old_rust_commit": "e457a7b0d326d67b4322ef0d11bd715cfaeda48f",
+    "generator_source_hashes": {
+      "probe.rs": "a2649d3ef94dcb7141ccabc435381009a555de5d67bfcf9c5295b99827ac4e89",
+      "render/probes.rs": "7eebba64372259c55f1fde06230a911e3930700d63f9ced486cf9d703207aba0",
+      "render/mod.rs": "aa1fdc47c98fa1430c0640ab2da6db574ad0a57179c0f1c79aca988729888a82",
+      "resolve/driver.rs": "2dad1922e2017535e81b7b4f0527c87dcc8e830243b8bc3ed852725f0f4eb65d"
+    },
+    "full_catalog_input_sha256": "016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7",
+    "candidate_catalog_inputs": {
+      "imported_sha256": "54d47b8e281dd953dc95fe16f834ff38164d9809171c9a1af097231b0bc45cab",
+      "overlay_sha256": "ecc7d0d7d70db737303abb6271666c67e3757a2727ba2191946a6d4e66e9f717",
+      "abi_ledger_sha256": "fc378e74ef6da50df31de62c3fafcb3264dfb414034e5852147359fcad4184eb",
+      "evidence_sha256": []
+    },
+    "primary_body_derivation": "Every primary input is byte-for-byte the corresponding committed intrinsics/probes/{id}.ll (or {id}.llvm_nvptx.ll), after restoring the catalog SHA-256 comment from captures column 7 and LLVM revision comment from primary_header_llvm_revisions. All 1699 inputs were reconstructed and hash-checked."
+  },
+  "primary_header_llvm_revisions": {
+    "primary_candidate": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+    "primary_full_catalog": "16696adcd119e6ba9cc175207d984d7021211acb"
+  },
+  "tables": {
+    "artifacts.jsonl": [
+      "sha256",
+      "bytes"
+    ],
+    "captures.jsonl": [
+      "intrinsic_id",
+      "mode_index",
+      "stage_index",
+      "codegen_gpu",
+      "ptx_feature",
+      "terminal_gpu_or_null",
+      "artifact_indices_by_kind",
+      "render_catalog_sha256_or_null",
+      "initial_candidate_exit_code"
+    ],
+    "stages.jsonl": [
+      "evidence_file_index",
+      "json_pointer",
+      "stage_index",
+      "ledger_sha256",
+      "proofs_as_capture_index_and_artifact_kind_index"
+    ],
+    "inputs.jsonl": [
+      "llvm_ir_artifact_index",
+      "exact_utf8_source"
+    ]
+  },
+  "indexing": "All indexes are zero-based. A JSONL row index is its zero-based line number. Null artifact indices mean that stage did not produce that artifact.",
+  "artifact_kinds": [
+    "llvm_ir",
+    "llvm_bitcode",
+    "canonical_llvm_ir",
+    "ptx",
+    "cubin"
+  ],
+  "stage_kinds": [
+    "all",
+    "declaration_canonicalization",
+    "backend_codegen",
+    "ptx_assembly"
+  ],
+  "capture_modes": [
+    "primary_candidate",
+    "primary_full_catalog",
+    "auxiliary_typed",
+    "auxiliary_inline"
+  ],
+  "evidence_files": [
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-clc.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-barrier.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-memory.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-sreg.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async-mbarrier.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-debug-control.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-elect.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-execution-control.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-extended-minmax.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-f32x2.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-integer-minmax.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-basic.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-extended.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8-f16x2.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-prmt.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-ampere-float.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-b1.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f16.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f32.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-fp8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m16.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8-m8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-mxf8f6f4-f32.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-arithmetic.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-conversion.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-math.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-fp8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8-m16n8k64.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-ampere-float.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f16.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f32.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k128.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k64.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8-m16n8k64.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-standard-int4.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-special-registers.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-stmatrix.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-control.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-cp.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ld.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ldst-offset.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-mma.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-st.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-threadfence.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-descriptor-control.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-reduce.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-wgmma-control.json",
+    "intrinsics/evidence/rust-llvm-23.1.3-aab11ee0.json"
+  ],
+  "original_profile_rename": {
+    "from": "rust-llvm-23.1.0-16696adc",
+    "to": "rust-llvm-23.1.3-aab11ee0"
+  },
+  "ledger_digest_recipe": "For each stage proof, dereference capture -> artifact index -> SHA-256. Sort triples (intrinsic_id, artifact_kind, SHA-256); SHA256 of concatenated UTF-8 lines \"{intrinsic_id} {artifact_kind} {sha256}\\n\" is ledger_sha256.",
+  "original_paths": {
+    "validation_root": "/tmp/cudaoxide-attr-20261011-validation",
+    "candidate_run_root": "/tmp/cudaoxide-intrinsics-rebaseline-20261011",
+    "primary_candidate": {
+      "original_prefix": "{candidate_run_root}/target/intrinsics/probes/{id}.candidate",
+      "suffixes": [
+        ".ll",
+        ".bc",
+        ".canonical.ll",
+        ".ptx",
+        ".cubin"
+      ],
+      "retained_prefix": "{validation_root}/intrinsics-primary-refresh/{id}/{codegen_gpu}-{ptx_without_plus}/{id}.candidate"
+    },
+    "primary_full_catalog": {
+      "render_prefix": "{validation_root}/intrinsics-primary-refresh/{id}/full-catalog-rendered/",
+      "render_names": [
+        "source.ll",
+        "{id}.bc",
+        "{id}.canonical.ll"
+      ],
+      "output_prefix": "{validation_root}/intrinsics-primary-refresh/{id}/{terminal_gpu_or_codegen_gpu}-{ptx_without_plus}/full-catalog/output",
+      "output_suffixes": [
+        ".ptx",
+        ".cubin"
+      ]
+    },
+    "auxiliary_typed": {
+      "prefix": "{validation_root}/auxiliary-typed/{id}.{stage_kind}{target_suffix}/{id}",
+      "target_suffix": "empty for declaration_canonicalization; otherwise .{codegen_gpu}.{ptx_without_plus}",
+      "suffixes": [
+        ".ll",
+        ".bc",
+        ".canonical.ll",
+        ".ptx",
+        ".cubin"
+      ]
+    },
+    "auxiliary_inline": {
+      "prefix": "{validation_root}/auxiliary-inline/{id}/",
+      "names": [
+        "source.ll",
+        null,
+        null,
+        "output.ptx",
+        "output.cubin"
+      ]
+    }
+  },
+  "compiler_commands": {
+    "llvm-as": [
+      "{llvm-as}",
+      "{original_llvm_ir_path}",
+      "-o",
+      "{original_llvm_bitcode_path}"
+    ],
+    "llvm-dis": [
+      "{llvm-dis}",
+      "{original_llvm_bitcode_path}",
+      "-o",
+      "{original_canonical_llvm_ir_path}"
+    ],
+    "llc": [
+      "{llc}",
+      "{original_llvm_ir_path}",
+      "-march=nvptx64",
+      "-mcpu={codegen_gpu}",
+      "-mattr={ptx_feature}",
+      "-o",
+      "{original_ptx_path}"
+    ],
+    "ptxas": [
+      "{ptxas}",
+      "-arch={terminal_gpu}",
+      "{original_ptx_path}",
+      "-o",
+      "{original_cubin_path}"
+    ]
+  },
+  "validation": {
+    "declarations": "Original crates/cuda-intrinsics-gen/src/probe.rs::assert_intrinsic_declaration_canonicalizes, including imported attribute checks.",
+    "ptx": "Original crates/cuda-intrinsics-gen/src/probe.rs::validate_probe_instructions, including exact pure instructions and selector checks.",
+    "capture_scaffolding": "Candidate CLI where supported; a temporary wrapper called original render::render_probe and probe validators for full-catalog fallback and auxiliary checks. The wrapper is not required to reproduce compilation: primary sources derive exactly from committed probes; every auxiliary source is retained here.",
+    "path_sensitive_artifacts": "llvm-as receives the original source path without a source_filename override; llvm-dis receives the original bitcode path. ModuleID/source_filename therefore affect recorded canonical IR and bitcode hashes. Restore original paths for byte-for-byte hash reproduction; use other paths for equivalent validation."
+  },
+  "extraction_commands": {
+    "checkout": [
+      "git",
+      "clone",
+      "https://github.com/rust-lang/llvm-project.git",
+      "{LLVM_SRC}"
+    ],
+    "revision": [
+      "git",
+      "-C",
+      "{LLVM_SRC}",
+      "checkout",
+      "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019"
+    ],
+    "configure": [
+      "cmake",
+      "-S",
+      "{LLVM_SRC}/llvm",
+      "-B",
+      "{LLVM_BUILD}",
+      "-G",
+      "Unix Makefiles",
+      "-DCMAKE_BUILD_TYPE=Release",
+      "-DLLVM_TARGETS_TO_BUILD=NVPTX",
+      "-DLLVM_ENABLE_PROJECTS=",
+      "-DLLVM_INCLUDE_TESTS=OFF",
+      "-DLLVM_INCLUDE_BENCHMARKS=OFF",
+      "-DLLVM_INCLUDE_EXAMPLES=OFF"
+    ],
+    "build": [
+      "cmake",
+      "--build",
+      "{LLVM_BUILD}",
+      "--target",
+      "llvm-tblgen",
+      "-j8"
+    ],
+    "intrinsics_dump": [
+      "{llvm-tblgen}",
+      "--dump-json",
+      "-I",
+      "{LLVM_SRC}/llvm/include",
+      "{LLVM_SRC}/llvm/include/llvm/IR/Intrinsics.td",
+      "-o",
+      "{intrinsics.json}"
+    ],
+    "nvptx_dump": [
+      "{llvm-tblgen}",
+      "--dump-json",
+      "-I",
+      "{LLVM_SRC}/llvm/include",
+      "-I",
+      "{LLVM_SRC}/llvm/lib/Target/NVPTX",
+      "{LLVM_SRC}/llvm/lib/Target/NVPTX/NVPTX.td",
+      "-o",
+      "{nvptx.json}"
+    ],
+    "normalize": [
+      "cargo",
+      "run",
+      "-p",
+      "cuda-intrinsics-gen",
+      "--",
+      "extract",
+      "--intrinsics-json",
+      "{intrinsics.json}",
+      "--nvptx-json",
+      "{nvptx.json}"
+    ]
+  },
+  "capture_checkout_commands": [
+    [
+      "git",
+      "clone",
+      "--shared",
+      "--no-checkout",
+      "/home/nihalp/prod/llvm-project",
+      "/tmp/cudaoxide-intrinsics-llvm-20261011"
+    ],
+    [
+      "git",
+      "-C",
+      "/tmp/cudaoxide-intrinsics-llvm-20261011",
+      "fetch",
+      "--depth=1",
+      "https://github.com/rust-lang/llvm-project.git",
+      "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019"
+    ],
+    [
+      "git",
+      "-C",
+      "/tmp/cudaoxide-intrinsics-llvm-20261011",
+      "checkout",
+      "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019"
+    ]
+  ],
+  "capture_extraction_paths": {
+    "LLVM_SRC": "/tmp/cudaoxide-intrinsics-llvm-20261011",
+    "LLVM_BUILD": "/tmp/cudaoxide-intrinsics-tblgen-build-20261011",
+    "intrinsics.json": "/tmp/cudaoxide-attr-20261011-validation/intrinsics-new-tblgen.json",
+    "nvptx.json": "/tmp/cudaoxide-attr-20261011-validation/intrinsics-new-nvptx-tblgen.json"
+  },
+  "checkout_note": "capture_checkout_commands records the local shared-checkout/fetch procedure; extraction_commands.checkout is a normalized public reproduction alternative. The configured generator and flags, compiler command argument order, and dump flags are the recorded capture settings.",
+  "table_hashes": {
+    "artifacts.jsonl": {
+      "sha256": "1f8b4b89f31500306c3db1cb1e0e8a665411b4df1bc68e08f3fb2781a0fccb11",
+      "bytes": 359939
+    },
+    "captures.jsonl": {
+      "sha256": "a0c3cc1e541399d2148fc1628ed1a738b8000a365503ac6c24c820d40fb0fe2d",
+      "bytes": 302046
+    },
+    "stages.jsonl": {
+      "sha256": "a399a1c3b5d7443631f60197cfe64201006a297701b6dbebb43b91d887462ef8",
+      "bytes": 107285
+    },
+    "inputs.jsonl": {
+      "sha256": "e42c4045df2a5113f3aee5e3ef560856f2ed73a4e5b11f1a48d0b76b1e7d6dc2",
+      "bytes": 112765
+    }
+  }
+}
diff --git a/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/stages.jsonl b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/stages.jsonl
new file mode 100644
index 0000000000..91b513949f
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/refresh-2026-10-11/stages.jsonl
@@ -0,0 +1,770 @@
+[0,"/fixtures/0/stages/0",1,"d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae",[[310,2],[320,2],[330,2],[340,2],[350,2],[360,2],[370,2],[380,2],[390,2],[400,2],[410,2],[420,2]]]
+[0,"/fixtures/0/stages/1",2,"6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028",[[310,3],[320,3],[330,3],[340,3],[350,3],[360,3],[370,3],[380,3],[390,3],[400,3],[410,3],[420,3]]]
+[0,"/fixtures/0/stages/2",2,"af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088",[[311,3],[321,3],[331,3],[341,3],[351,3],[361,3],[371,3],[381,3],[391,3],[401,3],[411,3],[421,3]]]
+[0,"/fixtures/0/stages/3",2,"566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7",[[312,3],[322,3],[332,3],[342,3],[352,3],[362,3],[372,3],[382,3],[392,3],[402,3],[412,3],[422,3]]]
+[0,"/fixtures/0/stages/4",2,"561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f",[[313,3],[323,3],[333,3],[343,3],[353,3],[363,3],[373,3],[383,3],[393,3],[403,3],[413,3],[423,3]]]
+[0,"/fixtures/0/stages/5",2,"49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8",[[314,3],[324,3],[334,3],[344,3],[354,3],[364,3],[374,3],[384,3],[394,3],[404,3],[414,3],[424,3]]]
+[0,"/fixtures/0/stages/6",2,"f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b",[[315,3],[325,3],[335,3],[345,3],[355,3],[365,3],[375,3],[385,3],[395,3],[405,3],[415,3],[425,3]]]
+[0,"/fixtures/0/stages/7",2,"b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532",[[316,3],[326,3],[336,3],[346,3],[356,3],[366,3],[376,3],[386,3],[396,3],[406,3],[416,3],[426,3]]]
+[0,"/fixtures/0/stages/8",2,"7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760",[[317,3],[327,3],[337,3],[347,3],[357,3],[367,3],[377,3],[387,3],[397,3],[407,3],[417,3],[427,3]]]
+[0,"/fixtures/0/stages/9",2,"2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63",[[318,3],[328,3],[338,3],[348,3],[358,3],[368,3],[378,3],[388,3],[398,3],[408,3],[418,3],[428,3]]]
+[0,"/fixtures/0/stages/10",2,"0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee",[[319,3],[329,3],[339,3],[349,3],[359,3],[369,3],[379,3],[389,3],[399,3],[409,3],[419,3],[429,3]]]
+[0,"/fixtures/0/stages/11",3,"df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6",[[310,4],[320,4],[330,4],[340,4],[350,4],[360,4],[370,4],[380,4],[390,4],[400,4],[410,4],[420,4]]]
+[0,"/fixtures/0/stages/12",3,"fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057",[[311,4],[321,4],[331,4],[341,4],[351,4],[361,4],[371,4],[381,4],[391,4],[401,4],[411,4],[421,4]]]
+[0,"/fixtures/0/stages/13",3,"717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada",[[312,4],[322,4],[332,4],[342,4],[352,4],[362,4],[372,4],[382,4],[392,4],[402,4],[412,4],[422,4]]]
+[0,"/fixtures/0/stages/14",3,"5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f",[[313,4],[323,4],[333,4],[343,4],[353,4],[363,4],[373,4],[383,4],[393,4],[403,4],[413,4],[423,4]]]
+[0,"/fixtures/0/stages/15",3,"0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f",[[314,4],[324,4],[334,4],[344,4],[354,4],[364,4],[374,4],[384,4],[394,4],[404,4],[414,4],[424,4]]]
+[0,"/fixtures/0/stages/16",3,"b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551",[[315,4],[325,4],[335,4],[345,4],[355,4],[365,4],[375,4],[385,4],[395,4],[405,4],[415,4],[425,4]]]
+[0,"/fixtures/0/stages/17",3,"6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5",[[316,4],[326,4],[336,4],[346,4],[356,4],[366,4],[376,4],[386,4],[396,4],[406,4],[416,4],[426,4]]]
+[0,"/fixtures/0/stages/18",3,"7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e",[[317,4],[327,4],[337,4],[347,4],[357,4],[367,4],[377,4],[387,4],[397,4],[407,4],[417,4],[427,4]]]
+[0,"/fixtures/0/stages/19",3,"c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc",[[318,4],[328,4],[338,4],[348,4],[358,4],[368,4],[378,4],[388,4],[398,4],[408,4],[418,4],[428,4]]]
+[0,"/fixtures/0/stages/20",3,"12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524",[[319,4],[329,4],[339,4],[349,4],[359,4],[369,4],[379,4],[389,4],[399,4],[409,4],[419,4],[429,4]]]
+[1,"/fixtures/0/stages/0",1,"49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e",[[58,2],[59,2],[60,2],[61,2],[62,2]]]
+[1,"/fixtures/0/stages/1",2,"90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316",[[58,3],[59,3],[60,3],[61,3],[62,3]]]
+[1,"/fixtures/0/stages/2",3,"0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2",[[58,4],[59,4],[60,4],[61,4],[62,4]]]
+[1,"/fixtures/1/stages/0",1,"bd29e45eadad739b58e43893296b4b8ac5e05d43c3c3298836fb126503d61e83",[[63,2]]]
+[1,"/fixtures/1/stages/1",2,"423445d4b9994ee20be92cbabc0ad7c8ea667c6360848df3b2a5be7020df4051",[[63,3]]]
+[1,"/fixtures/1/stages/2",3,"f8543fdf61a5a323f880af2fdfe1b86d2ee1311a024f5a9e88d6c40a23845925",[[63,4]]]
+[2,"/fixtures/0/stages/0",1,"709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22",[[33,2],[35,2],[39,2],[41,2]]]
+[2,"/fixtures/0/stages/1",2,"0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3",[[32,3],[34,3],[38,3],[40,3]]]
+[2,"/fixtures/0/stages/2",3,"7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37",[[32,4],[34,4],[38,4],[40,4]]]
+[2,"/fixtures/1/stages/0",1,"5c5b7ef9743a0c36a3ad27cf3093c5d3160111a0acee880922a93dab4ce330a7",[[36,2],[37,2]]]
+[2,"/fixtures/1/stages/1",2,"462bdc8a89623f986fa9c307a069d444805dcb8644ef7f9793f45a48a5cc94c6",[[36,3],[37,3]]]
+[2,"/fixtures/1/stages/2",3,"5501565e9a498f7f5dba887004bc2cf506c57249b8a4a2e8fdebcefc8007b146",[[36,4],[37,4]]]
+[3,"/fixtures/0/stages/0",2,"418a46a6a0279c0e2c7391fcccae3b3bec1a660a3ffad28b1df2a7f63df297d6",[[238,3],[438,3]]]
+[3,"/fixtures/0/stages/1",3,"bf4f5d63cf41eae708cc3087c76cbea8b9812ba26f1c915646ef4b0205c51760",[[238,4],[438,4]]]
+[4,"/fixtures/0/stages/0",1,"070eb640b89420af5d8d67631d8269d7b3efa14da6bad70274af322cf7c68ee4",[[68,2],[69,2],[70,2],[71,2],[72,2],[73,2],[74,2],[75,2],[76,2],[77,2],[78,2],[79,2],[80,2],[81,2]]]
+[4,"/fixtures/0/stages/1",2,"0059b9999da5a35149dd382b0eca010573789cfffa14abf3009e8bd4aa054eec",[[68,3],[69,3],[70,3],[71,3],[72,3],[73,3],[74,3],[75,3],[76,3],[77,3],[78,3],[79,3],[80,3],[81,3]]]
+[4,"/fixtures/0/stages/2",3,"06c87e1559d3544f2076dc749c8ca032775629f52dd5fe2995f0a93307534852",[[68,4],[69,4],[70,4],[71,4],[72,4],[73,4],[74,4],[75,4],[76,4],[77,4],[78,4],[79,4],[80,4],[81,4]]]
+[5,"/records/0/stages/0",1,"3d6f3f642d9e80245cb2f6c76442ccd0d824326e288898d121d5645368edcc8e",[[184,2]]]
+[5,"/records/0/stages/1",2,"8b4dfc81ee704ec2a8f8e8502b5f5673cafe35607eb28120b57fff0c3b27baf0",[[184,3]]]
+[5,"/records/0/stages/2",3,"86913ed12af6b0bd59e448e3b5ccc6823e2f0b6a140146ead578815ea6f7c5ec",[[184,4]]]
+[5,"/records/1/stages/0",1,"cd2056552d6e2b796c195f31ad768ca175eac447825d819f5957b9fcec07059c",[[187,2]]]
+[5,"/records/1/stages/1",2,"d59be5c8527b799671b85f003a16a51bbd07448cc2130863d25e7318f2841384",[[187,3]]]
+[5,"/records/1/stages/2",3,"23bdc05d4be1c27246a496b4f2ebede69efa8b84d326395adc4d804f16f9eef5",[[187,4]]]
+[5,"/records/2/stages/0",1,"11cd942627957bdbe07521415b591132bac6543a6d37a10286821ca4be498129",[[185,2]]]
+[5,"/records/2/stages/1",2,"d65ebc7c63cdfa57f0f5aefa206fd07f6004879beb44350640071d5eb97e6625",[[185,3]]]
+[5,"/records/2/stages/2",3,"8ef8193aee88f5efa25293d77c1399be0b83dd294401561177e91f144714072a",[[185,4]]]
+[5,"/records/3/stages/0",1,"5b62ad0b051d6ccaa435c3686d1a94547c3635490996a003e78abb772b160cc4",[[186,2]]]
+[5,"/records/3/stages/1",2,"9affe7627501776e21eb8059bdb4bee726a956435720cc6de05ee4d93ef51be1",[[186,3]]]
+[5,"/records/3/stages/2",3,"366217f18f9473cf5ccc1d87632a590d8fcf7a8f5508a5bf876c49df1dc939e2",[[186,4]]]
+[6,"/records/0/stages/0",1,"62367bcc29688f386037306dee01e279908da96b4fbc872a0731f6c1efe3b979",[[176,2]]]
+[6,"/records/0/stages/1",2,"fc90fb194ad06cc37e4ec41495d29b2ced1dadd23700f4a6742ba5cb274f236c",[[176,3]]]
+[6,"/records/0/stages/2",3,"61893d39abe94a66af4946061499f48731945913886e7c08b8e5c3761e2e2089",[[176,4]]]
+[6,"/records/1/stages/0",1,"7c8298238217b06520a14baa7dd9017ef6824f9a237ff1c1059c16bcf4a79797",[[179,2]]]
+[6,"/records/1/stages/1",2,"3022aaec048152a29d81d9b43ad7121669183c87def7ffd6bcc83d52d7040f91",[[179,3]]]
+[6,"/records/1/stages/2",3,"9928045f3f05dc39ca94c224c2afe7c21092310cdfa83a18da46628fa549c1cc",[[179,4]]]
+[6,"/records/2/stages/0",1,"b0555ffe6d212958fd651503471d80d0a62030dc35f1dbdac5aedd702be82dd9",[[177,2]]]
+[6,"/records/2/stages/1",2,"4115fad6c387b8a02ba11bff217ad8d8f244b96f080aad856ea45b4475ee27f8",[[177,3]]]
+[6,"/records/2/stages/2",3,"c2b95d90d47b38d58ceb6c07c46652eac5c8dec1e1ae2ede4f13972c17c1b268",[[177,4]]]
+[6,"/records/3/stages/0",1,"d0b171b000534f40633027c1b9b7ab26b739a7ebb686ec8617019f3834cad070",[[180,2]]]
+[6,"/records/3/stages/1",2,"a01e92a24c61a5a33a96cee208154315907b4538b97345be0e979839c5358c02",[[180,3]]]
+[6,"/records/3/stages/2",3,"aa2dd28dc4c740dac8fce5c55f0002a8bf7e9ab2210bacfd152777eb0e1928b0",[[180,4]]]
+[6,"/records/4/stages/0",1,"a4b7c596a91052b75958df78b963338ca733b9b53643eb4b669015b106f926a4",[[175,2]]]
+[6,"/records/4/stages/1",2,"99c87c9201c30820920f63624477dfab9046044be0c1e4e9e878a8bd5edd425f",[[175,3]]]
+[6,"/records/4/stages/2",3,"8d4109cb4c2deaedb98fce7f124afb56882fc7ca5ef239a53d86a293d152b465",[[175,4]]]
+[6,"/records/5/stages/0",1,"4eb30d0f094641ac8dc5af009c2818cf66555f890dc7e6ed4c8e92987f6ee493",[[178,2]]]
+[6,"/records/5/stages/1",2,"91e8999a4e6f9db5fd5ba08aa79bd58870a77e05e12b06523f182ec6cfb7a14d",[[178,3]]]
+[6,"/records/5/stages/2",3,"6d5bc6dde1a35b6d38bf05e0d3aea8d44520505a79dedfb9bc4fa3f96543e9c6",[[178,4]]]
+[6,"/records/6/stages/0",1,"f32b02e50561865fcbd41af2ab4e6d8c12485b436b14d191cbf5bbf0d248ea82",[[181,2]]]
+[6,"/records/6/stages/1",2,"f72afd215027bcf67444625672135ba3fbbfadbee3e28f8e90ced82b0321419e",[[181,3]]]
+[6,"/records/6/stages/2",3,"6e091384fbc0457e593996514ad9eeafc04773c58054a8bc85a8c96980c139e5",[[181,4]]]
+[6,"/records/7/stages/0",1,"ce9e52734dd2438c5fc127548069d9cb7130610dc9e0752f6598268db9a3a8fe",[[182,2]]]
+[6,"/records/7/stages/1",2,"5081c35375c4d19aaa528f8abbeb82c57237e9596018318466d212559d1dd65b",[[182,3]]]
+[6,"/records/7/stages/2",3,"3a0e1e9bc0608c3d6581f78c2e17d4ef3e71edcd3570f4bde0d9e8f545c8abd6",[[182,4]]]
+[6,"/records/8/stages/0",1,"9ea4574621afbb04281bfa8ab26db49b9a256ea5a991c63ff2907c9a725aef5a",[[183,2]]]
+[6,"/records/8/stages/1",2,"b2f1831e64e9e04bbe61a19f62e926eebb9e8c7a2e6e7f3fc073d97363c81f2a",[[183,3]]]
+[6,"/records/8/stages/2",3,"533d9b5b9a9c927d1fe0b23c76106771cdac35407624486b921bcf4c7096811a",[[183,4]]]
+[6,"/records/9/stages/0",1,"1b3dc9f7a7dd7ebd28ad446ba1b8a97a30ffa6065ab3b9bdbb2bcb4377287859",[[188,2]]]
+[6,"/records/9/stages/1",2,"2dfc842c70441fe57d5a892d34441cf37fedab3955f511c5ef23118632e06924",[[188,3]]]
+[6,"/records/9/stages/2",3,"96ab3b4af8cae0f8d415560129805aec75b9e3e8c26b916cda16a42cfc11d581",[[188,4]]]
+[6,"/records/10/stages/0",1,"c4b4d445968d5e8db78f64d6903e5dab71d7e0a7ed2edd66bff7dd5443900376",[[189,2]]]
+[6,"/records/10/stages/1",2,"0604968fa8260d2a46de521f458b9ea4da639127e0671340e531eb5af1e5bea6",[[189,3]]]
+[6,"/records/10/stages/2",3,"77ca31b604849a0fbb4a6eeff8a57d146053d576721be6dc2212ec7347f0f1c1",[[189,4]]]
+[7,"/fixtures/0/stages/0",2,"f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5",[[56,3],[1160,3],[1691,3]]]
+[7,"/fixtures/0/stages/1",3,"79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645",[[57,4],[1161,4],[1692,4]]]
+[8,"/fixtures/0/stages/0",1,"d6be8db0355a531bcc20af99d6b2ab9a8015c92ce61965522dec78614fd03d1c",[[241,2]]]
+[8,"/fixtures/0/stages/1",2,"de0e95f3672e5c96d15e4fd737d9225940692889c495211037f7b4f99d3cddb4",[[241,3]]]
+[8,"/fixtures/0/stages/2",3,"192c7ae696bf472409f022926d3c74f98784499ab59d769d1565e6605976c7d0",[[241,4]]]
+[9,"/records/0/stages/0",1,"1ec32a6b202b53ac34f270d293fd60d90d5d6de4dacef2b7d7912ab16698c69f",[[46,2]]]
+[9,"/records/0/stages/1",2,"fa2ed67b2503e249a9ae63ef2a7436c41d098c27956fe0faf441c0e961db4151",[[46,3]]]
+[9,"/records/0/stages/2",3,"8a9e71fa97f369891143a3c32fa8c80f1c28be09176445e1edfff81c54615ffd",[[47,4]]]
+[9,"/records/1/stages/0",1,"0c313db28069767d98c8d477dd5512e1d4dae29dc52ea0d691fe03edb03c5e50",[[48,2]]]
+[9,"/records/1/stages/1",2,"0615f66795bc5a379d6cf2426d83a39d4baae46194d61b6cde902e9b8c8ab367",[[48,3]]]
+[9,"/records/1/stages/2",3,"999e27bbd3d4ccd52c8f1366731b70ab35d9d67ed19125ffb66ac7bf160b4bbc",[[49,4]]]
+[9,"/records/2/stages/0",1,"70c603a8e3092012e74c6e6f5a2330489176c4af38499c379753f095ae899a2a",[[42,2]]]
+[9,"/records/2/stages/1",2,"6a153e4c537c56cc43402bcf7786c6540c696ff807919d11973d650c7bc9a78c",[[42,3]]]
+[9,"/records/2/stages/2",3,"9499277c4334862a892ac0c79eb49c79fc5c0dc889639a4b22dd5d0ea2134d0c",[[43,4]]]
+[9,"/records/3/stages/0",1,"171894b0dd863526a8ff4cf86ecc05f04a47032b49254ae9d5a03f58bcc5eaf4",[[44,2]]]
+[9,"/records/3/stages/1",2,"cb96f04a02a2ae0cde2c1d3691ca3d2752577a6b32337b39ab1afecaa7f98249",[[44,3]]]
+[9,"/records/3/stages/2",3,"11118e20cc0d2ec90152a19b6e75a9dbcc4f09a1d5698c030ebf890c88095fa1",[[45,4]]]
+[9,"/records/4/stages/0",1,"5b0abd382879e45371de563a83219bb7d4d07b5f9f42e1be22a6f4d677203678",[[297,2]]]
+[9,"/records/4/stages/1",2,"e1f0e1a7641d1458f0fa9d84525e830a2dbb84f58fab281009a922c356df6eec",[[297,3]]]
+[9,"/records/4/stages/2",3,"306ce9690dd2d976940425ec7052d1f0cdb3b0cee03e8507345d328dde684796",[[297,4]]]
+[9,"/records/5/stages/0",1,"0e63a89f7a0e5ef9325815f030775cc2f52a995704e695e7f6f6bc0a9d29e7e2",[[298,2]]]
+[9,"/records/5/stages/1",2,"351547f724188c103868fd7cf963ba564b216c62db91f2a2a24a8685feb892a6",[[298,3]]]
+[9,"/records/5/stages/2",3,"10bee9dcaa765674caa7cfcc9059aa263e5cf7b95fcb8fbdca3bee1f03e2b589",[[298,4]]]
+[9,"/records/6/stages/0",1,"c6648953f506cce13f94612bf6d0da7cd3e075fbc63a6afc01f2ae709fb9a415",[[1211,2]]]
+[9,"/records/6/stages/1",2,"4032051215c86d96bb88330e799aec22fb09ad3d232fe483e2346692108ff967",[[1211,3]]]
+[9,"/records/6/stages/2",3,"06608466abf4fe979cff6b7e5e3a16a92d0daf5df071239d28e943cd7be29916",[[1211,4]]]
+[9,"/records/7/stages/0",1,"be9be58f6dcddd987a8632ca75a2ed6a592f6a65ceff960adc24bcb52bb11db3",[[1210,2]]]
+[9,"/records/7/stages/1",2,"ddfec7d31a47990bd677ada57d94afc3bf77325e79e886219aac161fffc0f896",[[1210,3]]]
+[9,"/records/7/stages/2",3,"37ad4f1284701ac5101858b0792ed16e833420d4e21cbc1c02896d03019a63a2",[[1210,4]]]
+[10,"/fixtures/0/stages/0",1,"26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a",[[1770,2],[1776,2],[1800,2],[1806,2],[1848,2],[1854,2],[1878,2],[1884,2]]]
+[10,"/fixtures/0/stages/1",2,"5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a",[[1771,3],[1777,3],[1801,3],[1807,3],[1849,3],[1855,3],[1879,3],[1885,3]]]
+[10,"/fixtures/0/stages/2",3,"c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5",[[1772,4],[1778,4],[1802,4],[1808,4],[1850,4],[1856,4],[1880,4],[1886,4]]]
+[10,"/fixtures/0/stages/3",2,"6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098",[[448,3],[450,3],[464,3],[466,3],[507,3],[509,3],[523,3],[525,3]]]
+[10,"/fixtures/0/stages/4",3,"c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5",[[448,4],[450,4],[464,4],[466,4],[507,4],[509,4],[523,4],[525,4]]]
+[10,"/fixtures/3/stages/0",1,"27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f",[[1782,2],[1785,2],[1791,2],[1794,2],[1812,2],[1818,2],[1821,2],[1827,2],[1833,2],[1836,2],[1860,2],[1863,2],[1869,2],[1872,2],[1890,2],[1896,2],[1899,2],[1905,2],[1911,2],[1914,2]]]
+[10,"/fixtures/3/stages/1",2,"b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec",[[1783,3],[1786,3],[1792,3],[1795,3],[1813,3],[1819,3],[1822,3],[1828,3],[1834,3],[1837,3],[1861,3],[1864,3],[1870,3],[1873,3],[1891,3],[1897,3],[1900,3],[1906,3],[1912,3],[1915,3]]]
+[10,"/fixtures/3/stages/2",3,"f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e",[[1784,4],[1787,4],[1793,4],[1796,4],[1814,4],[1820,4],[1823,4],[1829,4],[1835,4],[1838,4],[1862,4],[1865,4],[1871,4],[1874,4],[1892,4],[1898,4],[1901,4],[1907,4],[1913,4],[1916,4]]]
+[10,"/fixtures/3/stages/3",2,"0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b",[[453,3],[455,3],[459,3],[461,3],[469,3],[473,3],[475,3],[483,3],[487,3],[489,3],[512,3],[514,3],[518,3],[520,3],[528,3],[532,3],[534,3],[542,3],[546,3],[548,3]]]
+[10,"/fixtures/3/stages/4",3,"f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e",[[453,4],[455,4],[459,4],[461,4],[469,4],[473,4],[475,4],[483,4],[487,4],[489,4],[512,4],[514,4],[518,4],[520,4],[528,4],[532,4],[534,4],[542,4],[546,4],[548,4]]]
+[10,"/fixtures/1/stages/0",1,"fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda",[[1761,2],[1764,2],[1767,2],[1773,2],[1797,2],[1803,2],[1839,2],[1842,2],[1845,2],[1851,2],[1875,2],[1881,2]]]
+[10,"/fixtures/1/stages/1",2,"9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684",[[1762,3],[1765,3],[1768,3],[1774,3],[1798,3],[1804,3],[1840,3],[1843,3],[1846,3],[1852,3],[1876,3],[1882,3]]]
+[10,"/fixtures/1/stages/2",3,"b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026",[[1763,4],[1766,4],[1769,4],[1775,4],[1799,4],[1805,4],[1841,4],[1844,4],[1847,4],[1853,4],[1877,4],[1883,4]]]
+[10,"/fixtures/1/stages/3",2,"e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39",[[443,3],[445,3],[447,3],[449,3],[463,3],[465,3],[502,3],[504,3],[506,3],[508,3],[522,3],[524,3]]]
+[10,"/fixtures/1/stages/4",3,"b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026",[[443,4],[445,4],[447,4],[449,4],[463,4],[465,4],[502,4],[504,4],[506,4],[508,4],[522,4],[524,4]]]
+[10,"/fixtures/2/stages/0",1,"f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1",[[1779,2],[1788,2],[1809,2],[1815,2],[1824,2],[1830,2],[1857,2],[1866,2],[1887,2],[1893,2],[1902,2],[1908,2]]]
+[10,"/fixtures/2/stages/1",2,"76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa",[[1780,3],[1789,3],[1810,3],[1816,3],[1825,3],[1831,3],[1858,3],[1867,3],[1888,3],[1894,3],[1903,3],[1909,3]]]
+[10,"/fixtures/2/stages/2",3,"8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87",[[1781,4],[1790,4],[1811,4],[1817,4],[1826,4],[1832,4],[1859,4],[1868,4],[1889,4],[1895,4],[1904,4],[1910,4]]]
+[10,"/fixtures/2/stages/3",2,"91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474",[[451,3],[457,3],[467,3],[471,3],[481,3],[485,3],[510,3],[516,3],[526,3],[530,3],[540,3],[544,3]]]
+[10,"/fixtures/2/stages/4",3,"8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87",[[451,4],[457,4],[467,4],[471,4],[481,4],[485,4],[510,4],[516,4],[526,4],[530,4],[540,4],[544,4]]]
+[11,"/fixtures/0/stages/0",2,"28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1",[[7,3],[8,3],[264,3],[267,3],[1134,3],[1135,3],[1261,3],[1262,3]]]
+[11,"/fixtures/0/stages/1",3,"b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487",[[7,4],[8,4],[264,4],[267,4],[1134,4],[1135,4],[1261,4],[1262,4]]]
+[12,"/records/0/stages/0",2,"05737e13d7ecd4d7d29377123087680401a96d0f6423842f9eb0c570c819a156",[[537,3]]]
+[12,"/records/1/stages/0",2,"47ca916b21f4f81062a57a273ed3a2574d88dc7961f3ed061ee81135c02019a6",[[478,3]]]
+[12,"/records/2/stages/0",2,"2b9b8cd02f5dbb4b21bbe79e50ef96e87c4ff8a61d113774a049c1df97e2d645",[[538,3]]]
+[12,"/records/3/stages/0",2,"019612a6c0a895437ac9e29439ca4404a44adf617072fb390270e41680557112",[[479,3]]]
+[12,"/records/4/stages/0",2,"bc038f0362d6863309339980bbb161649c4774ba6fc3e43058645c320efd1724",[[539,3]]]
+[12,"/records/5/stages/0",2,"493b5e74e07a0232486d2df12ed0c80ed712745b36b51c17cfec678c28fdcb97",[[480,3]]]
+[12,"/records/6/stages/0",2,"ccd77090590ed5077784a28895d4db44c0c3601913263b77b3afedf890a94db1",[[536,3]]]
+[12,"/records/7/stages/0",2,"e5fc39021e5a2dd08e1b4c600bd65bfbd719061fd7182b4f87325793bcfeca3a",[[477,3]]]
+[13,"/records/0/stages/0",1,"28fcac92a1ebd5fca5408ed34c7aa64ba3546bbe5714355a7ebfd14b92f760e3",[[496,2]]]
+[13,"/records/0/stages/1",2,"50ba004fcfb85da96f6f3347966e4704dda4443ace6b57676924bd528c08bf0b",[[496,3]]]
+[13,"/records/0/stages/2",3,"8ab45efc26f5f14c513e6ab0ed925ffddbdfc1cb0e890ef4f445a40e88acebc9",[[496,4]]]
+[13,"/records/1/stages/0",1,"7df70d42d790f340f00c6f29da571b9b809bee9f1d093b742ce1a0e421272f46",[[491,2]]]
+[13,"/records/1/stages/1",2,"8d140364a2d31493758ea19f5e725bbfea20f14d6ce43489b079bb4789e428ce",[[491,3]]]
+[13,"/records/1/stages/2",3,"57bdb80f49dc0b09308813d01303ede4a2ba34e99e63aaecd6bea6813137fd7e",[[491,4]]]
+[13,"/records/2/stages/0",1,"3eb1f951460f6bc67c6055899542860e1010e5db5406f5e2f9accee3cbc170c2",[[495,2]]]
+[13,"/records/2/stages/1",2,"174b9c4a382274b14d72edd3fd22bdc8d2486f5cb87d6c8bb4a48e7bc66b7000",[[495,3]]]
+[13,"/records/2/stages/2",3,"b579f9834ca9bfb229788f02fc0ce145ba59fac61cbb61036bc8c552cb4bbb7a",[[495,4]]]
+[13,"/records/3/stages/0",2,"c6a81cac5ab0bf8f94c54775467e1df856112351608671f11a4e8e63ffb75f00",[[498,3]]]
+[13,"/records/3/stages/1",3,"df51ed0e79da4ce10d9298aa7e4fc3db4b75355044d51f95f92694acde7f9e50",[[498,4]]]
+[13,"/records/4/stages/0",1,"76483bcaa272bf6c2c1688ff61ce630eba2cbc651af55d047b7048e43352adb7",[[497,2]]]
+[13,"/records/4/stages/1",2,"4effb270bc9d6ccd5d7c24a4518d532eb913541a234db702e0d50b7c4f7f0774",[[497,3]]]
+[13,"/records/4/stages/2",3,"453f943cafeb60e5761ed5f0867ebb1e2519a3dd314eb9b0dc4b5e6ff6fbe65c",[[497,4]]]
+[14,"/fixtures/1/stages/0",2,"43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c",[[249,3],[252,3],[492,3],[493,3],[501,3]]]
+[14,"/fixtures/1/stages/1",3,"c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0",[[249,4],[252,4],[492,4],[493,4],[501,4]]]
+[14,"/fixtures/2/stages/0",2,"ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97",[[250,3],[251,3],[494,3]]]
+[14,"/fixtures/2/stages/1",3,"383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b",[[250,4],[251,4],[494,4]]]
+[14,"/fixtures/0/stages/0",2,"5dbfe0da451289673522a20ec37fd0664d5276bf7760b0d8a8484d613e82ca01",[[499,3],[500,3]]]
+[14,"/fixtures/0/stages/1",3,"24bb1e758b565ffc3df774b43f1cdc6f7b26a5f827d1add384e6ffcd78747080",[[499,4],[500,4]]]
+[14,"/fixtures/3/stages/0",2,"bb1f49e5307d0e617ee6cf95da7ec7a582167ac87e14e8bc4d24cb4185ee7f68",[[1148,3]]]
+[14,"/fixtures/3/stages/1",3,"68ce336a8ad5452c31fbff9377bccbcb5540d81a83b904ec1cf2641f607fb34a",[[1149,4]]]
+[15,"/records/0/stages/0",1,"777bd7a5362a935cb9b45c99eb5c7912d8296fd322f232be5feb4dcb71afe958",[[1733,2]]]
+[15,"/records/0/stages/1",2,"8843a1e5b9fee11892601ac9688d35a5195bdd8c88122bb6ab5b4fa21cf76e29",[[1734,3]]]
+[15,"/records/0/stages/2",3,"ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb",[[1735,4]]]
+[15,"/records/0/stages/3",2,"8d344820320e9db759a6db9eebd370929159ce71d483a76cb70f846d27abf426",[[200,3]]]
+[15,"/records/0/stages/4",3,"ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb",[[200,4]]]
+[15,"/records/1/stages/0",1,"a17f8309a9b1436ceae14d34278cbc25993c1607850ec87962d7fef9fd6c3215",[[1739,2]]]
+[15,"/records/1/stages/1",2,"11284b3464e72006408b832095a07a2ee8683336b367d675109a20b5289db239",[[1740,3]]]
+[15,"/records/1/stages/2",3,"086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7",[[1741,4]]]
+[15,"/records/1/stages/3",2,"82c1a28789b05b8df6d92cc52932afdc99dc761cb2bcc77c7d85e2b3a1da1e24",[[204,3]]]
+[15,"/records/1/stages/4",3,"086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7",[[204,4]]]
+[15,"/records/2/stages/0",1,"5b6fcb8ea52dc533221fc3a953835ab9fb2cdb8a3c82a1dc9527dc2b980cb096",[[1736,2]]]
+[15,"/records/2/stages/1",2,"a6707ccb77a07e8b004032798bbfd07afba3af30d15568ac2c3d08d84d8b10b3",[[1737,3]]]
+[15,"/records/2/stages/2",3,"f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a",[[1738,4]]]
+[15,"/records/2/stages/3",2,"6064f89491e6f84f9ac56c6373eec145a9b8d55507b30a4ee2ffcbe2deed7dd1",[[202,3]]]
+[15,"/records/2/stages/4",3,"f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a",[[202,4]]]
+[15,"/records/3/stages/0",1,"2a8ce890aff0fa2df126aba76cf0dcc3fbda114de759ad215043e6ad0d400d23",[[1742,2]]]
+[15,"/records/3/stages/1",2,"ab076ae9d2de65d0ab3caad272a118bd426726141914f2b539a30a616d655e4e",[[1743,3]]]
+[15,"/records/3/stages/2",3,"8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2",[[1744,4]]]
+[15,"/records/3/stages/3",2,"cf21d0dd82dfc3a278f16fc60cf6772892bc2faee051ae9dd92b23f6e934f314",[[206,3]]]
+[15,"/records/3/stages/4",3,"8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2",[[206,4]]]
+[15,"/records/4/stages/0",1,"b295a2577d8a238ef3098f7cb0e4fc309f9e5f1177f9f02ae766101f9d70cc67",[[1715,2]]]
+[15,"/records/4/stages/1",2,"fdceb080a053c25340804b87375d5fc8693a5fc8eaa461399d438982c1cd179a",[[1716,3]]]
+[15,"/records/4/stages/2",3,"5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9",[[1717,4]]]
+[15,"/records/4/stages/3",2,"5682cacdfec0f3e7fe02db10b1665b3b88ab176b4db26738c965155f31ba0dbf",[[192,3]]]
+[15,"/records/4/stages/4",3,"5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9",[[192,4]]]
+[15,"/records/5/stages/0",1,"e12dd4763bbc1533bd40687021b8319bd843af8e62fc1fefe8212335310449ab",[[1724,2]]]
+[15,"/records/5/stages/1",2,"e0264fce6d34633b3225bd3f48723a6efaa11b18a738168e34c460dfe2b16754",[[1725,3]]]
+[15,"/records/5/stages/2",3,"9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e",[[1726,4]]]
+[15,"/records/5/stages/3",2,"8a2341dac9ffd42fede2aefd66f576cac3e464c0b6b81c38773128832579eafa",[[195,3]]]
+[15,"/records/5/stages/4",3,"9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e",[[195,4]]]
+[15,"/records/6/stages/0",1,"8a572f8ac9426c4acabd244748fbcabfffc19a7afbc3eb6b74b087a7397d0ca5",[[1718,2]]]
+[15,"/records/6/stages/1",2,"182d55254c86ae42a4eaaf4db2c9829f771eb0f23f6abaa92f773fd9910b607b",[[1719,3]]]
+[15,"/records/6/stages/2",3,"52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68",[[1720,4]]]
+[15,"/records/6/stages/3",2,"e7d96da22340431d9ae279f1ea4d4573bca35a171e7fc6199b5893b540a93424",[[193,3]]]
+[15,"/records/6/stages/4",3,"52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68",[[193,4]]]
+[15,"/records/7/stages/0",1,"b32d14ca011f505a526f0fecf85660bd93e1bbc147e728896df311e7f2b1176b",[[1727,2]]]
+[15,"/records/7/stages/1",2,"f69dc7503144fa82436f92c37e77254ebf11ca61ffdcf492ce8b6ace6ae00234",[[1728,3]]]
+[15,"/records/7/stages/2",3,"41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33",[[1729,4]]]
+[15,"/records/7/stages/3",2,"dff12609bfb2cbd5258c5d20e2e1c8b781090475d157dfe239cdbfc5edd9c204",[[196,3]]]
+[15,"/records/7/stages/4",3,"41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33",[[196,4]]]
+[16,"/fixtures/0/stages/0",1,"bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab",[[201,2],[203,2],[205,2],[207,2]]]
+[16,"/fixtures/0/stages/1",2,"fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a",[[201,3],[203,3],[205,3],[207,3]]]
+[16,"/fixtures/0/stages/2",3,"a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2",[[201,4],[203,4],[205,4],[207,4]]]
+[16,"/fixtures/0/stages/3",2,"9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c",[[2068,3],[2070,3],[2072,3],[2074,3]]]
+[16,"/fixtures/0/stages/4",3,"a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2",[[2069,4],[2071,4],[2073,4],[2075,4]]]
+[17,"/fixtures/0/stages/0",1,"a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2",[[1163,2],[1165,2],[1167,2],[1169,2],[1171,2],[1173,2],[1175,2]]]
+[17,"/fixtures/0/stages/1",2,"69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2",[[1163,3],[1165,3],[1167,3],[1169,3],[1171,3],[1173,3],[1175,3]]]
+[17,"/fixtures/0/stages/2",3,"30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72",[[1164,4],[1166,4],[1168,4],[1170,4],[1172,4],[1174,4],[1176,4]]]
+[18,"/fixtures/0/stages/0",1,"7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a",[[1917,2],[2034,2],[2035,2],[2036,2],[2037,2]]]
+[18,"/fixtures/0/stages/1",2,"680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe",[[552,3],[897,3],[906,3],[907,3],[908,3]]]
+[18,"/fixtures/0/stages/2",3,"979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4",[[552,4],[897,4],[906,4],[907,4],[908,4]]]
+[19,"/records/0/stages/0",2,"666182ec743c9ffe0422975ab4ab4faf2e3ddcff656913f34303f15200fa3700",[[911,3]]]
+[19,"/records/0/stages/1",3,"452782b4303bcdb3357c76efeae34f913d781f5673cd5a97eace443e1dc28b77",[[911,4]]]
+[19,"/records/1/stages/0",2,"4fc3a76fa4f4f3118a8078863963902f0b33ebeecb269fccefaf2d785a9171f4",[[551,3]]]
+[19,"/records/1/stages/1",3,"66efb066dd43977add4c7b6812754fc0affb5166a6dee22b060759d4eac2e5ab",[[551,4]]]
+[19,"/records/2/stages/0",2,"e760db526d0a3aef6cdfb3a2032edb8ef925f23d7c046b15eb455e304c0a66b2",[[572,3]]]
+[19,"/records/2/stages/1",3,"f577f397dc588d57cf9709716a198f6b17f1d4f6fa071c808bcaff2ddb58b8dc",[[572,4]]]
+[19,"/records/3/stages/0",2,"c6cf34f19543565ab3b0a268c1b38020771261bc358df589709391ec5e99ad3f",[[910,3]]]
+[19,"/records/3/stages/1",3,"93107316bf0cf5ef84b40e52ec5ef683b2781a3e3a0e75737e4892184d0c6457",[[910,4]]]
+[19,"/records/4/stages/0",2,"0786eb2ac48a149064adcb908b142249e47b70b7dc6b77221c691534d54f8363",[[550,3]]]
+[19,"/records/4/stages/1",3,"d7868cbe80c20cad2dafd27dc6429792a50fd3b4efb0da2dd55b4d836fedaef9",[[550,4]]]
+[19,"/records/5/stages/0",2,"b08bb4ac9ef5d95f902982d6d6e4da9bc6e758f02a86e4f0da412b6dd22f561c",[[571,3]]]
+[19,"/records/5/stages/1",3,"fd92d37a566e4ee028ef24ef3bc175b98a0ade64ebf874e29c45a409119ee2bb",[[571,4]]]
+[20,"/fixtures/0/stages/0",1,"3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c",[[1926,2],[1928,2],[1930,2],[1932,2],[1934,2],[1936,2],[1938,2],[1940,2],[1942,2],[1944,2],[1946,2],[1948,2],[1950,2],[1952,2],[1954,2],[1956,2],[1958,2],[1960,2],[1962,2],[1964,2],[1966,2],[1968,2],[1970,2],[1972,2],[1974,2]]]
+[20,"/fixtures/0/stages/1",2,"5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e",[[1927,3],[1929,3],[1931,3],[1933,3],[1935,3],[1937,3],[1939,3],[1941,3],[1943,3],[1945,3],[1947,3],[1949,3],[1951,3],[1953,3],[1955,3],[1957,3],[1959,3],[1961,3],[1963,3],[1965,3],[1967,3],[1969,3],[1971,3],[1973,3],[1975,3]]]
+[20,"/fixtures/0/stages/2",2,"4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b",[[573,3],[577,3],[581,3],[585,3],[589,3],[593,3],[597,3],[601,3],[605,3],[609,3],[613,3],[617,3],[621,3],[625,3],[629,3],[633,3],[637,3],[641,3],[645,3],[649,3],[653,3],[657,3],[661,3],[665,3],[669,3]]]
+[20,"/fixtures/0/stages/3",3,"1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2",[[573,4],[577,4],[581,4],[585,4],[589,4],[593,4],[597,4],[601,4],[605,4],[609,4],[613,4],[617,4],[621,4],[625,4],[629,4],[633,4],[637,4],[641,4],[645,4],[649,4],[653,4],[657,4],[661,4],[665,4],[669,4]]]
+[20,"/fixtures/0/stages/4",2,"8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3",[[574,3],[578,3],[582,3],[586,3],[590,3],[594,3],[598,3],[602,3],[606,3],[610,3],[614,3],[618,3],[622,3],[626,3],[630,3],[634,3],[638,3],[642,3],[646,3],[650,3],[654,3],[658,3],[662,3],[666,3],[670,3]]]
+[20,"/fixtures/0/stages/5",3,"89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8",[[574,4],[578,4],[582,4],[586,4],[590,4],[594,4],[598,4],[602,4],[606,4],[610,4],[614,4],[618,4],[622,4],[626,4],[630,4],[634,4],[638,4],[642,4],[646,4],[650,4],[654,4],[658,4],[662,4],[666,4],[670,4]]]
+[20,"/fixtures/0/stages/6",2,"c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e",[[575,3],[579,3],[583,3],[587,3],[591,3],[595,3],[599,3],[603,3],[607,3],[611,3],[615,3],[619,3],[623,3],[627,3],[631,3],[635,3],[639,3],[643,3],[647,3],[651,3],[655,3],[659,3],[663,3],[667,3],[671,3]]]
+[20,"/fixtures/0/stages/7",3,"a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc",[[575,4],[579,4],[583,4],[587,4],[591,4],[595,4],[599,4],[603,4],[607,4],[611,4],[615,4],[619,4],[623,4],[627,4],[631,4],[635,4],[639,4],[643,4],[647,4],[651,4],[655,4],[659,4],[663,4],[667,4],[671,4]]]
+[20,"/fixtures/0/stages/8",2,"b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de",[[576,3],[580,3],[584,3],[588,3],[592,3],[596,3],[600,3],[604,3],[608,3],[612,3],[616,3],[620,3],[624,3],[628,3],[632,3],[636,3],[640,3],[644,3],[648,3],[652,3],[656,3],[660,3],[664,3],[668,3],[672,3]]]
+[20,"/fixtures/0/stages/9",3,"40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d",[[576,4],[580,4],[584,4],[588,4],[592,4],[596,4],[600,4],[604,4],[608,4],[612,4],[616,4],[620,4],[624,4],[628,4],[632,4],[636,4],[640,4],[644,4],[648,4],[652,4],[656,4],[660,4],[664,4],[668,4],[672,4]]]
+[21,"/fixtures/0/stages/0",1,"dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d",[[1976,2],[1978,2],[1980,2],[1982,2],[1984,2],[1986,2],[1988,2],[1990,2],[1992,2],[1994,2],[1996,2],[1998,2],[2000,2],[2002,2],[2004,2],[2006,2],[2008,2],[2010,2],[2012,2],[2014,2],[2016,2],[2018,2],[2020,2],[2022,2],[2024,2]]]
+[21,"/fixtures/0/stages/1",2,"3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f",[[1977,3],[1979,3],[1981,3],[1983,3],[1985,3],[1987,3],[1989,3],[1991,3],[1993,3],[1995,3],[1997,3],[1999,3],[2001,3],[2003,3],[2005,3],[2007,3],[2009,3],[2011,3],[2013,3],[2015,3],[2017,3],[2019,3],[2021,3],[2023,3],[2025,3]]]
+[21,"/fixtures/0/stages/2",2,"21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c",[[673,3],[677,3],[681,3],[685,3],[689,3],[693,3],[697,3],[701,3],[705,3],[709,3],[713,3],[717,3],[721,3],[725,3],[729,3],[733,3],[737,3],[741,3],[745,3],[749,3],[753,3],[757,3],[761,3],[765,3],[769,3]]]
+[21,"/fixtures/0/stages/3",3,"d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa",[[673,4],[677,4],[681,4],[685,4],[689,4],[693,4],[697,4],[701,4],[705,4],[709,4],[713,4],[717,4],[721,4],[725,4],[729,4],[733,4],[737,4],[741,4],[745,4],[749,4],[753,4],[757,4],[761,4],[765,4],[769,4]]]
+[21,"/fixtures/0/stages/4",2,"9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e",[[674,3],[678,3],[682,3],[686,3],[690,3],[694,3],[698,3],[702,3],[706,3],[710,3],[714,3],[718,3],[722,3],[726,3],[730,3],[734,3],[738,3],[742,3],[746,3],[750,3],[754,3],[758,3],[762,3],[766,3],[770,3]]]
+[21,"/fixtures/0/stages/5",3,"4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa",[[674,4],[678,4],[682,4],[686,4],[690,4],[694,4],[698,4],[702,4],[706,4],[710,4],[714,4],[718,4],[722,4],[726,4],[730,4],[734,4],[738,4],[742,4],[746,4],[750,4],[754,4],[758,4],[762,4],[766,4],[770,4]]]
+[21,"/fixtures/0/stages/6",2,"f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451",[[675,3],[679,3],[683,3],[687,3],[691,3],[695,3],[699,3],[703,3],[707,3],[711,3],[715,3],[719,3],[723,3],[727,3],[731,3],[735,3],[739,3],[743,3],[747,3],[751,3],[755,3],[759,3],[763,3],[767,3],[771,3]]]
+[21,"/fixtures/0/stages/7",3,"8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e",[[675,4],[679,4],[683,4],[687,4],[691,4],[695,4],[699,4],[703,4],[707,4],[711,4],[715,4],[719,4],[723,4],[727,4],[731,4],[735,4],[739,4],[743,4],[747,4],[751,4],[755,4],[759,4],[763,4],[767,4],[771,4]]]
+[21,"/fixtures/0/stages/8",2,"8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab",[[676,3],[680,3],[684,3],[688,3],[692,3],[696,3],[700,3],[704,3],[708,3],[712,3],[716,3],[720,3],[724,3],[728,3],[732,3],[736,3],[740,3],[744,3],[748,3],[752,3],[756,3],[760,3],[764,3],[768,3],[772,3]]]
+[21,"/fixtures/0/stages/9",3,"a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274",[[676,4],[680,4],[684,4],[688,4],[692,4],[696,4],[700,4],[704,4],[708,4],[712,4],[716,4],[720,4],[724,4],[728,4],[732,4],[736,4],[740,4],[744,4],[748,4],[752,4],[756,4],[760,4],[764,4],[768,4],[772,4]]]
+[22,"/fixtures/0/stages/0",1,"ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e",[[1918,2],[1919,2],[1920,2],[1921,2]]]
+[22,"/fixtures/0/stages/1",2,"fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61",[[555,3],[556,3],[557,3],[558,3]]]
+[22,"/fixtures/0/stages/2",3,"0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db",[[555,4],[556,4],[557,4],[558,4]]]
+[22,"/fixtures/1/stages/0",1,"400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746",[[1922,2],[1923,2],[1924,2],[1925,2]]]
+[22,"/fixtures/1/stages/1",2,"2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8",[[559,3],[560,3],[561,3],[562,3]]]
+[22,"/fixtures/1/stages/2",3,"6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2",[[559,4],[560,4],[561,4],[562,4]]]
+[22,"/fixtures/2/stages/0",1,"789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55",[[2026,2],[2027,2],[2028,2],[2029,2]]]
+[22,"/fixtures/2/stages/1",2,"727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9",[[773,3],[774,3],[775,3],[776,3]]]
+[22,"/fixtures/2/stages/2",3,"71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b",[[773,4],[774,4],[775,4],[776,4]]]
+[22,"/fixtures/3/stages/0",1,"259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f",[[2030,2],[2031,2],[2032,2],[2033,2]]]
+[22,"/fixtures/3/stages/1",2,"69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84",[[777,3],[778,3],[779,3],[780,3]]]
+[22,"/fixtures/3/stages/2",3,"858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a",[[777,4],[778,4],[779,4],[780,4]]]
+[23,"/records/0/stages/0",2,"c47aa6dd3e24383098f128a64c87c47ab4797f5e033e3fc4bb0a670d2503013a",[[881,3]]]
+[23,"/records/0/stages/1",3,"c3aa0206bc409a773e92374c337ae29700fa54735792eeaafe1d8f87fcb9c523",[[881,4]]]
+[23,"/records/1/stages/0",2,"ab9e908618637e0d0699f72f1a3a80b33f56c41e57a5c802af03819ef2c5ef14",[[883,3]]]
+[23,"/records/1/stages/1",3,"7567be8747a7b101aa8df77028eef647926a3d0dea2f0fcd83d3ebe953a622e2",[[883,4]]]
+[23,"/records/2/stages/0",2,"9cd0a0a84b94f37267419e33805b31860d5685fcd2cf4564783e28343a972748",[[889,3]]]
+[23,"/records/2/stages/1",3,"f2bac6e9f7ce9e6cb01846a046917c140500e7e876942100e7d3ce977cbfd90d",[[889,4]]]
+[23,"/records/3/stages/0",2,"a51cb813a45b9cd0c0e9dc7038bec77a14194732f4b180ec96eecd3958d6a169",[[890,3]]]
+[23,"/records/3/stages/1",3,"3b2c092a2e9c78fd3580523dc6cc875e8903497e790eee28cfa15324c03685a7",[[890,4]]]
+[23,"/records/4/stages/0",2,"d9b3d3e50e66a2281bd4a40ade4118f0afb4d7789eef88afc79e9498ac5a3ea2",[[882,3]]]
+[23,"/records/4/stages/1",3,"8ca1b55fd6cb49b5380574ff42ec7d55b6dbf557c08ff72873a5da7eecd90838",[[882,4]]]
+[23,"/records/5/stages/0",2,"81ae38ebf4bd8a59ff18a33203383a22cace76e383f473b31a6debf0f9ee947f",[[884,3]]]
+[23,"/records/5/stages/1",3,"973c539e9cb7d01605f08ede31f1bcf9fb12155ae80857ea6da2f5bf085509f6",[[884,4]]]
+[23,"/records/6/stages/0",2,"023ff2fe53a4cd8bb534791bde0e7384b5820f1fb7eb9cb1bed3627b8bed4c7e",[[892,3]]]
+[23,"/records/6/stages/1",3,"fedf24465346088c564e7971f6b7c29e92fceb052adb100fafb91b5e5b3b6bee",[[892,4]]]
+[23,"/records/7/stages/0",2,"ce1cd3a7682dc726d32cf180d1b39bdf9ac3aea8f12aa2a6c6423b7f521a15ab",[[891,3]]]
+[23,"/records/7/stages/1",3,"a5085e6a26e349f53eba46f4e630ecd2142786cbed096ed927acc94cc9c5d910",[[891,4]]]
+[23,"/records/8/stages/0",2,"b5ff5fcfbf7af82721582a788d144fe6745bd1ea783aef2ff9102a328612ce3e",[[898,3]]]
+[23,"/records/8/stages/1",3,"95520ee8bd9137e246a8f6b10d90bf414fd6640b6941209df067082273bab6a4",[[898,4]]]
+[23,"/records/9/stages/0",2,"a4b6f9f1214b02bafdfe07cca12072fdbfab865362fd3b295e688a738be8d0df",[[900,3]]]
+[23,"/records/9/stages/1",3,"814a279d3913788ca9cc088325b84198c97e9693aaa13778c28f0f654b834e38",[[900,4]]]
+[23,"/records/10/stages/0",2,"48af8df444ba8fcff6c181e3862bbcfabfeb9817b970ddb66ea95df74814252e",[[902,3]]]
+[23,"/records/10/stages/1",3,"e7e02b52b0a757ec780b5e93f59ff1abdea9770a82d4f0f4d1384035db9f803b",[[902,4]]]
+[23,"/records/11/stages/0",2,"c9140abf7847a4b891673bba722aaef6a63ce9520ae3764efd411de6d636208f",[[903,3]]]
+[23,"/records/11/stages/1",3,"3824b5037253381a849ded485ed000a3c459f68a94dcd43b809a72958bf618aa",[[903,4]]]
+[23,"/records/12/stages/0",2,"5b42070729512dc4a34d959933cb2ed579c549ca2c4a34ba53634c793c2104c5",[[899,3]]]
+[23,"/records/12/stages/1",3,"052fb66d0981ed2361e6cde56d62007088995b7ba713d6da1691e0012e7f4923",[[899,4]]]
+[23,"/records/13/stages/0",2,"c6d117f7dc3771deff164ddacd83b64a80d1b9c3f33543c2323394f860296b3a",[[901,3]]]
+[23,"/records/13/stages/1",3,"232ee3454e7542b49e296aafcd4c9f38235fe2141fb224ca3f329aec36d456b3",[[901,4]]]
+[23,"/records/14/stages/0",2,"59332065be960c742528cb03bad200c2c5a5ad472285420dd46116d3d9da8d88",[[905,3]]]
+[23,"/records/14/stages/1",3,"2b6ed03aae00d63ec51d849f1382eb95b8580686067589bea8eaa9433e3d8be1",[[905,4]]]
+[23,"/records/15/stages/0",2,"fa313e124cffc714546231b65e670c63303446e7edb95dce4af94ca57bf258a3",[[904,3]]]
+[23,"/records/15/stages/1",3,"3ed63a17572d63f68c5f40f43cf5920d3432723ec6b946bd1689e3c44d3fe4f9",[[904,4]]]
+[24,"/records/0/stages/0",2,"242a7d2104a977a3c344c947eb5c03d10a18711cb776b10a27c2519d3760eeb0",[[920,3]]]
+[24,"/records/0/stages/1",3,"c5b6ee75c034c9af33c18ab6c91f725b42d2b54c8bf132ec1b89c115cb9a3548",[[920,4]]]
+[24,"/records/1/stages/0",2,"57e936cd723c8522e8d7ad7c20eec7f1efe473d9dde5381bc74874bbd737acc1",[[922,3]]]
+[24,"/records/1/stages/1",3,"540c2a2b1c2c2945022dad6f87d3cdf948ade2eab7a1aa2149e47b320adf4e48",[[922,4]]]
+[24,"/records/2/stages/0",2,"16ea31bd434789327320d0683d1cc83a11058292365655503c77c2143139e1e4",[[924,3]]]
+[24,"/records/2/stages/1",3,"75676b7cb4e4fb123537b9f8f1cc40ccb04e1db0271ab3eb760961f4ea73ea37",[[924,4]]]
+[24,"/records/3/stages/0",2,"a777e8a890e81f7bc799e08ffa31129ff28d46f13cd9bbb9467f7d3e09e294b6",[[925,3]]]
+[24,"/records/3/stages/1",3,"87501053789841d181fd1aae8dac5b4dc516639d99983f1867fd6f4df9be4836",[[925,4]]]
+[24,"/records/4/stages/0",2,"b5fb1110c752bf4fe08f37bd8b7f8345d676ca3eea504e71bed250fb3c7497ce",[[921,3]]]
+[24,"/records/4/stages/1",3,"5e9c1003612b5e1b0441919a9d51eeadbd9a9c26f365be978cf5d279b0b0d1fc",[[921,4]]]
+[24,"/records/5/stages/0",2,"1a0ad568976eb0b009eeb8941f96e74169d76cd8f5e486c761848b16a6a78f68",[[923,3]]]
+[24,"/records/5/stages/1",3,"ee342c6e9f3f578c32c6af5dd1c2668cf556669258abfcce6f85298f77a2fe90",[[923,4]]]
+[24,"/records/6/stages/0",2,"9b77eb2760494d673de880beb0e9da2110a690897b8968ba0b33d57e0f7b635f",[[927,3]]]
+[24,"/records/6/stages/1",3,"666fe69ffd45aab6a3f5bfc5183c41460a862957228ada97aee5dece4d4b92a4",[[927,4]]]
+[24,"/records/7/stages/0",2,"d3a2ccb17359a5a84b5f2ffbbbf21bdbf3e862d6a2a6e007cba7442aa65144a9",[[926,3]]]
+[24,"/records/7/stages/1",3,"787d75bc93a0a84ac830fa9dd7472adc96e9b78362d7e6ce1b107301d5c6d0a6",[[926,4]]]
+[25,"/records/0/stages/0",2,"b5c2da0b56637465b6318014f3bb75f68d843c14ef827e70a7ef937625ad5e2c",[[912,3]]]
+[25,"/records/0/stages/1",3,"df22078cac5b20809b7d5a9eea6a29d6cfe354500d77fa360d86a0b4a1023852",[[912,4]]]
+[25,"/records/1/stages/0",2,"ab1d12b93c4b140169d3169350a16500008aa4163f09614a631b7e1dc4fdbcb1",[[914,3]]]
+[25,"/records/1/stages/1",3,"a423b06a09e13590b1d54c6f8d8be51d1e394145354b004a635293e1d76ac83c",[[914,4]]]
+[25,"/records/2/stages/0",2,"4f6d577d18158476e7ab49d17a2ad0f47da8b6ae752a6ed080d20e1380d52bee",[[916,3]]]
+[25,"/records/2/stages/1",3,"1955df76beb6ee81ed7ea879c3200f4f6bb7e1966deb92e945b3f325841dc5cd",[[916,4]]]
+[25,"/records/3/stages/0",2,"6f75f6988bc1dda288d7c571d72462c4298a17b37aaa830a033de583852ac772",[[917,3]]]
+[25,"/records/3/stages/1",3,"4dff1215835b5b26444fd0e25f94789efea21a227da12d555c889c796c0ea93f",[[917,4]]]
+[25,"/records/4/stages/0",2,"270630238dd5c5f9cf517a25b950096e73e12e4479f81ac8c1cd3283017390d0",[[913,3]]]
+[25,"/records/4/stages/1",3,"1bf5493b1e849d63814126e394e6f0572529d702a2433d0de1ff47b7b4ac0650",[[913,4]]]
+[25,"/records/5/stages/0",2,"d10485a6ecf25b1ef218ba5ca3641e58f8066afb750fe46d83566b3fe2b4f072",[[915,3]]]
+[25,"/records/5/stages/1",3,"2f5aaa3aac438dbac75e659a7a874e86e99e3f0114d38a53f3d889f2063ea0b6",[[915,4]]]
+[25,"/records/6/stages/0",2,"51db7a6fe697ef26e445bac4abdaa4d5e45e887e8aa5a366f608654d3466aab0",[[919,3]]]
+[25,"/records/6/stages/1",3,"06aafc80a8ca8190b809819eaac9c4776807f6380a05ed5a02ce479898506506",[[919,4]]]
+[25,"/records/7/stages/0",2,"93bd9995092da3436c05eb1720405ceeea64728cf2eb687f907c7b1c820d0148",[[918,3]]]
+[25,"/records/7/stages/1",3,"67dbf19d3684c67bd2b2ffface224d08f9e24a216e617ebf5307e530bc492d97",[[918,4]]]
+[26,"/records/0/stages/0",2,"a48e36d7da44b51913001f1eb7d4d1a4c3fe187828db3a37451079cfb79bd041",[[563,3]]]
+[26,"/records/0/stages/1",3,"3f2318b8382d87d1f67a2cc34a6842dbb8d910a85130f0d3d991690b39d6f1c2",[[563,4]]]
+[26,"/records/1/stages/0",2,"c7ad5689bb1dc7d9f6e66a3b7024246cecd620c8499b9468999c5efc7cf4ef38",[[565,3]]]
+[26,"/records/1/stages/1",3,"2318600ab58b33347303547db32bfbfdccea5210d1401891255d10c18a8d89cd",[[565,4]]]
+[26,"/records/2/stages/0",2,"db2d33329053739e1f39831d56e2d27e487dd334679bb09de2f5c8f80eb1de65",[[567,3]]]
+[26,"/records/2/stages/1",3,"0a674a423572b924a7f2e6dd0a2c69d601e5d2c4682dfa1414103e7045a686c6",[[567,4]]]
+[26,"/records/3/stages/0",2,"9921a009519c555a2adb72baf0e136405ef6b850b9a19d83f10e7d0b3d763f71",[[568,3]]]
+[26,"/records/3/stages/1",3,"969755d83267f71bbb72953caff716f8eae65ee1d14c4a293879a0a21338c337",[[568,4]]]
+[26,"/records/4/stages/0",2,"77110080515c64834cb555469a0ea69ed12c499b30be821f8832fdcfac0f73c5",[[887,3]]]
+[26,"/records/4/stages/1",3,"7700bbce0406dc6977a0af2702a6679b1a41e88c867eed3903621b9d14beef53",[[887,4]]]
+[26,"/records/5/stages/0",2,"67ac1fcd13f430b3a14dc6d8467cc9beec1a0b175c1e785fce2a9a91545d379d",[[893,3]]]
+[26,"/records/5/stages/1",3,"0ba09c1ddf924a125f7171df66dec143a9c4d598837bdaed8de694fc7b52b4b8",[[893,4]]]
+[26,"/records/6/stages/0",2,"ec734a83add935db593a1f90090b395a74af19c9c84661e61a80d46ba07966ca",[[894,3]]]
+[26,"/records/6/stages/1",3,"13d684b6400fcb97db851b1a163797e7b6cabd462f847f3f54b3342a0d49697c",[[894,4]]]
+[26,"/records/7/stages/0",2,"32f364dd111daf1d85405149c434903c9ba832ee2d5d6e3067ab468e61bcd713",[[564,3]]]
+[26,"/records/7/stages/1",3,"f136aaf2137b3c1f962fdcdbeaf7f7ed4ad688be4fb5cd55dadaf7ac100fc153",[[564,4]]]
+[26,"/records/8/stages/0",2,"516ff5a0c66113c514e59c0272876c73360c509581521ac8ef9c6c7b1a02c82e",[[566,3]]]
+[26,"/records/8/stages/1",3,"b15a9cd586a5fbde549415441be60eae06e57d99604ba2350ca476c0066f6a1a",[[566,4]]]
+[26,"/records/9/stages/0",2,"332115a18b93fa8d2c7916c9fe028636dde30234d0b197244146f55eca052148",[[570,3]]]
+[26,"/records/9/stages/1",3,"8301b6cd488dea73b3278090a9a30910809f2afba5c3e9b58b21a1d84cdca8bd",[[570,4]]]
+[26,"/records/10/stages/0",2,"ab5b0e911466d87855d48864c54ff2da53c2e4e5b940b1c0812b378ea453e585",[[569,3]]]
+[26,"/records/10/stages/1",3,"a300eed5eea4deeaf14e77ce212087a11d0a912a110956a8f98d7ef0219018cf",[[569,4]]]
+[26,"/records/11/stages/0",2,"c512bacbd7f6da60ef9eda11b065823f600b9f0409a0ca562d8b2ae0e2c2df48",[[886,3]]]
+[26,"/records/11/stages/1",3,"57e571ed96c84be6aaa7c9428a16c4c002f791b55429f427820fcc3d6fbb177c",[[886,4]]]
+[26,"/records/12/stages/0",2,"16e7d991475a0e8577e129c8337f9eb2ba08ec792a3500cd9890a5be58311b2b",[[888,3]]]
+[26,"/records/12/stages/1",3,"74351d962b7eb0ac0ca27e606b8d6e75468b39752b667b28e6277475de18bb29",[[888,4]]]
+[26,"/records/13/stages/0",2,"eaabf735e8ebc1b813e38c3ec0d3c20f167786603c98ce8ebc1867abdf0e5d52",[[896,3]]]
+[26,"/records/13/stages/1",3,"deefb4ba049125c5b91b39ac823909f05e24c773ceac2520b10210d11e02c17b",[[896,4]]]
+[26,"/records/14/stages/0",2,"f2c069f3d47afb6653954d8f58fb146b9e927136d703104a92a2b8312c874d45",[[895,3]]]
+[26,"/records/14/stages/1",3,"fffbf71c999883fa695a7300a982460fc7e6425a322521cc9e98523dcfdc3fc2",[[895,4]]]
+[27,"/fixtures/0/stages/0",2,"4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872",[[781,3],[785,3],[789,3],[793,3],[797,3],[801,3],[805,3],[809,3],[813,3],[817,3],[821,3],[825,3],[829,3],[833,3],[837,3],[841,3],[845,3],[849,3],[853,3],[857,3],[861,3],[865,3],[869,3],[873,3],[877,3]]]
+[27,"/fixtures/0/stages/1",3,"e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1",[[781,4],[785,4],[789,4],[793,4],[797,4],[801,4],[805,4],[809,4],[813,4],[817,4],[821,4],[825,4],[829,4],[833,4],[837,4],[841,4],[845,4],[849,4],[853,4],[857,4],[861,4],[865,4],[869,4],[873,4],[877,4]]]
+[27,"/fixtures/0/stages/2",2,"5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25",[[782,3],[786,3],[790,3],[794,3],[798,3],[802,3],[806,3],[810,3],[814,3],[818,3],[822,3],[826,3],[830,3],[834,3],[838,3],[842,3],[846,3],[850,3],[854,3],[858,3],[862,3],[866,3],[870,3],[874,3],[878,3]]]
+[27,"/fixtures/0/stages/3",3,"cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9",[[782,4],[786,4],[790,4],[794,4],[798,4],[802,4],[806,4],[810,4],[814,4],[818,4],[822,4],[826,4],[830,4],[834,4],[838,4],[842,4],[846,4],[850,4],[854,4],[858,4],[862,4],[866,4],[870,4],[874,4],[878,4]]]
+[27,"/fixtures/0/stages/4",2,"79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e",[[783,3],[787,3],[791,3],[795,3],[799,3],[803,3],[807,3],[811,3],[815,3],[819,3],[823,3],[827,3],[831,3],[835,3],[839,3],[843,3],[847,3],[851,3],[855,3],[859,3],[863,3],[867,3],[871,3],[875,3],[879,3]]]
+[27,"/fixtures/0/stages/5",3,"1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3",[[783,4],[787,4],[791,4],[795,4],[799,4],[803,4],[807,4],[811,4],[815,4],[819,4],[823,4],[827,4],[831,4],[835,4],[839,4],[843,4],[847,4],[851,4],[855,4],[859,4],[863,4],[867,4],[871,4],[875,4],[879,4]]]
+[27,"/fixtures/0/stages/6",2,"8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27",[[784,3],[788,3],[792,3],[796,3],[800,3],[804,3],[808,3],[812,3],[816,3],[820,3],[824,3],[828,3],[832,3],[836,3],[840,3],[844,3],[848,3],[852,3],[856,3],[860,3],[864,3],[868,3],[872,3],[876,3],[880,3]]]
+[27,"/fixtures/0/stages/7",3,"fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77",[[784,4],[788,4],[792,4],[796,4],[800,4],[804,4],[808,4],[812,4],[816,4],[820,4],[824,4],[828,4],[832,4],[836,4],[840,4],[844,4],[848,4],[852,4],[856,4],[860,4],[864,4],[868,4],[872,4],[876,4],[880,4]]]
+[28,"/records/0/stages/0",2,"0a3bd41a37b501ef0419e036d4a2325f28dee3d5d47c050ab1af6d52b9904385",[[553,3]]]
+[28,"/records/0/stages/1",3,"5b0d9dbc95cc97fb422f1b7b9e39d1d68c866b4ffcf48c9a32f8fb917454bb82",[[553,4]]]
+[28,"/records/1/stages/0",2,"cb21825affcdfcb250dea8bfead8827d200376f1b6dfb69b7659d670b4fe0a8d",[[554,3]]]
+[28,"/records/1/stages/1",3,"d25700d38dead0336c44add15c4dc07020aab017e285b96e0e94bf091e743e4e",[[554,4]]]
+[28,"/records/2/stages/0",2,"229d640dcb7295b409e4d690f6094a704af0b77e69e834abb5c2c9d26ac5daa0",[[909,3]]]
+[28,"/records/2/stages/1",3,"7543fed70ec348791089d56f1eee6d51e13501a1a40092f118bec2c99b734d3c",[[909,4]]]
+[28,"/records/3/stages/0",2,"6524c9c1c5d31e51a7bc4f69ed88bb6f64bc9718c4cd4e6dbdf38d71cceb7406",[[885,3]]]
+[28,"/records/3/stages/1",3,"f67ea6f8fc76d2142c8297e2fbef29257d5a89eba91380d4ebbbf47cdce91489",[[885,4]]]
+[28,"/records/4/stages/0",2,"455ca35747a458536fae20c94d1749e6b7c07dc9b482fb469192376c9e10a1a8",[[928,3]]]
+[28,"/records/4/stages/1",3,"f7bf4efce75e805867c6ef541acdd239b2df1f6819376e1d08b24cf76398b507",[[928,4]]]
+[29,"/fixtures/0/stages/0",1,"43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d",[[218,2],[219,2],[220,2],[221,2],[222,2],[223,2],[224,2],[225,2],[226,2],[227,2],[228,2],[229,2],[273,2],[274,2],[275,2],[278,2],[279,2],[280,2],[283,2],[284,2],[285,2],[288,2],[289,2],[290,2],[1136,2],[1137,2],[1138,2],[1139,2],[1140,2],[1141,2],[1142,2],[1143,2],[1144,2],[1145,2],[1146,2],[1147,2]]]
+[29,"/fixtures/0/stages/1",2,"3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2",[[218,3],[219,3],[220,3],[221,3],[222,3],[223,3],[224,3],[225,3],[226,3],[227,3],[228,3],[229,3],[273,3],[274,3],[275,3],[278,3],[279,3],[280,3],[283,3],[284,3],[285,3],[288,3],[289,3],[290,3],[1136,3],[1137,3],[1138,3],[1139,3],[1140,3],[1141,3],[1142,3],[1143,3],[1144,3],[1145,3],[1146,3],[1147,3]]]
+[29,"/fixtures/0/stages/2",3,"33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb",[[218,4],[219,4],[220,4],[221,4],[222,4],[223,4],[224,4],[225,4],[226,4],[227,4],[228,4],[229,4],[273,4],[274,4],[275,4],[278,4],[279,4],[280,4],[283,4],[284,4],[285,4],[288,4],[289,4],[290,4],[1136,4],[1137,4],[1138,4],[1139,4],[1140,4],[1141,4],[1142,4],[1143,4],[1144,4],[1145,4],[1146,4],[1147,4]]]
+[29,"/fixtures/2/stages/0",1,"f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954",[[1699,2],[1700,2],[1701,2],[1702,2],[1703,2],[1704,2],[1705,2],[1706,2],[1751,2],[1752,2],[1753,2],[1754,2],[1755,2],[1756,2],[1757,2],[1758,2]]]
+[29,"/fixtures/2/stages/1",2,"e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336",[[12,3],[13,3],[17,3],[18,3],[22,3],[23,3],[27,3],[28,3],[276,3],[277,3],[281,3],[282,3],[286,3],[287,3],[291,3],[292,3]]]
+[29,"/fixtures/2/stages/2",3,"584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0",[[12,4],[13,4],[17,4],[18,4],[22,4],[23,4],[27,4],[28,4],[276,4],[277,4],[281,4],[282,4],[286,4],[287,4],[291,4],[292,4]]]
+[29,"/fixtures/1/stages/0",1,"9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d",[[9,2],[10,2],[11,2],[14,2],[15,2],[16,2],[19,2],[20,2],[21,2],[24,2],[25,2],[26,2]]]
+[29,"/fixtures/1/stages/1",2,"3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807",[[9,3],[10,3],[11,3],[14,3],[15,3],[16,3],[19,3],[20,3],[21,3],[24,3],[25,3],[26,3]]]
+[29,"/fixtures/1/stages/2",3,"0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64",[[9,4],[10,4],[11,4],[14,4],[15,4],[16,4],[19,4],[20,4],[21,4],[24,4],[25,4],[26,4]]]
+[30,"/fixtures/0/stages/0",1,"7fc68746b0ee106bcb3270f8a02cb83cd0a4e85363c0854f82546717a33bf72d",[[211,2]]]
+[30,"/fixtures/0/stages/1",2,"f5355197dbd45ff46716deac34ba4d8de1084ee4374f537af9a67dd2fb47b6c5",[[211,3]]]
+[30,"/fixtures/0/stages/2",3,"8bf2670d9bea4e04e196ffe05f2954a5d9f8beefc57a231af094444b20348c3a",[[211,4]]]
+[30,"/fixtures/1/stages/0",1,"8a38ab8cd30065c8554fc11f7b6f155740eb533d290525b57a45a4b339465234",[[210,2]]]
+[30,"/fixtures/1/stages/1",2,"e2819c3052033ee04b85149f6891347ea860e7bacc511a0e7edb1d5fb4f36cd2",[[210,3]]]
+[30,"/fixtures/1/stages/2",3,"e8c2a92a43cdb367ea7856ca832f6fea44f45613b70f89f2acd2b45c4d8b35b4",[[210,4]]]
+[30,"/fixtures/2/stages/0",1,"f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203",[[199,2],[209,2],[215,2],[217,2]]]
+[30,"/fixtures/2/stages/1",2,"b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a",[[199,3],[209,3],[215,3],[217,3]]]
+[30,"/fixtures/2/stages/2",3,"36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937",[[199,4],[209,4],[215,4],[217,4]]]
+[30,"/fixtures/3/stages/0",1,"3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71",[[198,2],[208,2],[214,2],[216,2]]]
+[30,"/fixtures/3/stages/1",2,"cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb",[[198,3],[208,3],[214,3],[216,3]]]
+[30,"/fixtures/3/stages/2",3,"5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c",[[198,4],[208,4],[214,4],[216,4]]]
+[31,"/fixtures/1/stages/0",1,"848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082",[[1707,2],[1708,2],[1759,2],[1760,2],[2063,2],[2064,2],[2065,2],[2066,2],[2067,2]]]
+[31,"/fixtures/1/stages/1",2,"094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b",[[82,3],[83,3],[436,3],[437,3],[1207,3],[1208,3],[1209,3],[1236,3],[1237,3]]]
+[31,"/fixtures/1/stages/2",3,"832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540",[[82,4],[83,4],[436,4],[437,4],[1207,4],[1208,4],[1209,4],[1236,4],[1237,4]]]
+[31,"/fixtures/0/stages/0",1,"fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125",[[1177,2],[1178,2],[1179,2],[1180,2],[1181,2],[1182,2],[1183,2],[1184,2],[1185,2],[1186,2],[1187,2],[1188,2],[1189,2],[1190,2],[1240,2],[1241,2],[1242,2],[1243,2],[1244,2],[1245,2],[1246,2],[1247,2],[1248,2],[1249,2],[1250,2],[1251,2],[1252,2],[1253,2]]]
+[31,"/fixtures/0/stages/1",2,"872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536",[[1177,3],[1178,3],[1179,3],[1180,3],[1181,3],[1182,3],[1183,3],[1184,3],[1185,3],[1186,3],[1187,3],[1188,3],[1189,3],[1190,3],[1240,3],[1241,3],[1242,3],[1243,3],[1244,3],[1245,3],[1246,3],[1247,3],[1248,3],[1249,3],[1250,3],[1251,3],[1252,3],[1253,3]]]
+[31,"/fixtures/0/stages/2",3,"40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a",[[1177,4],[1178,4],[1179,4],[1180,4],[1181,4],[1182,4],[1183,4],[1184,4],[1185,4],[1186,4],[1187,4],[1188,4],[1189,4],[1190,4],[1240,4],[1241,4],[1242,4],[1243,4],[1244,4],[1245,4],[1246,4],[1247,4],[1248,4],[1249,4],[1250,4],[1251,4],[1252,4],[1253,4]]]
+[31,"/fixtures/3/stages/0",2,"cee684a507daecda40b46b40a8de9a2f2582dfcf5aae88b3d844b49e614c44e7",[[246,3]]]
+[31,"/fixtures/3/stages/1",3,"0fd052b8dd9f1821ac995eba3175320422af7c36dcdf0fc99190a1b2819de6e6",[[246,4]]]
+[31,"/fixtures/2/stages/0",2,"7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb",[[247,3],[248,3],[1267,3]]]
+[31,"/fixtures/2/stages/1",3,"d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a",[[247,4],[248,4],[1267,4]]]
+[32,"/fixtures/0/stages/0",2,"b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047",[[945,3],[946,3],[947,3],[948,3]]]
+[32,"/fixtures/0/stages/1",3,"9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4",[[945,4],[946,4],[947,4],[948,4]]]
+[33,"/records/0/stages/0",2,"fcd1b6080f311a524c25afacc87c7db3e52597ff2da06494ecc63a319740ff3a",[[953,3]]]
+[33,"/records/0/stages/1",3,"e7c617c1291bbecb8b6aa72f56558797c56ea20bb7a78fd5769aae5d0f12b720",[[953,4]]]
+[33,"/records/1/stages/0",2,"f9dcb133309ca12ab7c5fabd0703c014103473203370652bf89f8b5c39de9898",[[955,3]]]
+[33,"/records/1/stages/1",3,"b95b07cec0fa97e908c16d96d9ded6697b9173ef6debe699b291aa890a5d22ee",[[955,4]]]
+[33,"/records/2/stages/0",2,"73597b682d1b12c2fccfad62f40ece013c3c3fcd9024e59c32ef84a8732364e5",[[961,3]]]
+[33,"/records/2/stages/1",3,"ac63519d007e8dce250d4517871606b252765146f7f59abdd1602790cb331543",[[961,4]]]
+[33,"/records/3/stages/0",2,"c4d93f3a3ddbc1b78370fc6d41ef8514a6a7aabaa7ec338b788f32dd08f910ae",[[962,3]]]
+[33,"/records/3/stages/1",3,"46b4764da009f8939b74a6235f29b9a6fed892e8b86f68ff56d576e2e5d11151",[[962,4]]]
+[33,"/records/4/stages/0",2,"f1627fbb5b81a5df542173347c1dabb3e3493c579f2d1ec962fbfb8bf88936a7",[[954,3]]]
+[33,"/records/4/stages/1",3,"207e717b6d247cfb221fdbcec7a626708ce5b740bdb21f409d1dbbb439941c5e",[[954,4]]]
+[33,"/records/5/stages/0",2,"35ffdccb3d2495814cf57898071b7b916a6b6ca5627fa62e179523389496a18d",[[956,3]]]
+[33,"/records/5/stages/1",3,"619c473b81d7936bbd80cd8b1195a5911b756c6209ce85beba9dc3b63c929df0",[[956,4]]]
+[33,"/records/6/stages/0",2,"a3f0063a7b3cacd62f63384b92d8719e8a4697c12df321ff9a1d5dece0e84d13",[[964,3]]]
+[33,"/records/6/stages/1",3,"ff140019dd06530e11bd9507f0e371f07c5d88300c8dbbcd619681437d175825",[[964,4]]]
+[33,"/records/7/stages/0",2,"ee679b12117e5c9162717c0d88a751b96da7576d08049d06c19e3fb6c6300c73",[[963,3]]]
+[33,"/records/7/stages/1",3,"00174476515d29940f2a3bd1efce00363d2005a85306c962ce53c6a9bc84e43b",[[963,4]]]
+[34,"/records/0/stages/0",2,"6117422b739d9c39ae85c36a064b6ba0cfe6242cbd6c19fffe7ff68365e00a79",[[937,3]]]
+[34,"/records/0/stages/1",3,"a84718234f047484ec8d86c72e2a333a2982ea25b4266f109070266baa49c715",[[937,4]]]
+[34,"/records/1/stages/0",2,"32f1c01ee1a43ca81812d076ee8db48b8928aa441daf3f71dfc025f6b50a516a",[[939,3]]]
+[34,"/records/1/stages/1",3,"875de648ec423e0f311608024d48bd81ae45ae442c9741a8f08c01502f863e9b",[[939,4]]]
+[34,"/records/2/stages/0",2,"34822f024cc535508a1d7ae0c22eb25c61a337e07e674e4c5d0821c85d2e44e0",[[941,3]]]
+[34,"/records/2/stages/1",3,"c0cad4768d71256792c04fdf7e962f3630275a62a909d6ecd15a67a1aacb7765",[[941,4]]]
+[34,"/records/3/stages/0",2,"1dc739b2f8f0bc7f5d2a744a638f9f9e93a9c9763584bdc4ef239e3b740a7900",[[942,3]]]
+[34,"/records/3/stages/1",3,"11793a16a1edc4a91afc40987b925e8829ac10289034931cf23a5dd19c54ff17",[[942,4]]]
+[34,"/records/4/stages/0",2,"3c5d9bb6256f7009234918046b2dfb11507265c220722e1a42aa20653a409aec",[[938,3]]]
+[34,"/records/4/stages/1",3,"9c0a5d19a7e1637ff335da9a214e99e06463d5160df778e1bad3dfcfc6874890",[[938,4]]]
+[34,"/records/5/stages/0",2,"b79bb4f92abfb0ce7a3b13baed56134e8043e73d4552d305d42f8f2d97661bfd",[[940,3]]]
+[34,"/records/5/stages/1",3,"adbd36486601138ffe021f493473ae535351c523380e780a11678e1a4da0e4e7",[[940,4]]]
+[34,"/records/6/stages/0",2,"d8c7f3a382ee9d1ea29abdb323d012394df4bf6bce1fa24e8594c78e8bfa65f7",[[944,3]]]
+[34,"/records/6/stages/1",3,"46586ec2c81c1e15a80b74485e9e14cad94c67af27deead4a2ce433787a814af",[[944,4]]]
+[34,"/records/7/stages/0",2,"d51f3473017b1bc3eafcae85544debb79cd50a621baebf76491730e6af4dfd83",[[943,3]]]
+[34,"/records/7/stages/1",3,"c8b4cd0226a362e5b29976f9f58f7fba85de7a0f25858de6e77bef814660815e",[[943,4]]]
+[35,"/fixtures/0/stages/0",2,"bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5",[[973,3],[974,3],[975,3],[976,3],[977,3],[978,3],[979,3],[1129,3]]]
+[35,"/fixtures/0/stages/1",3,"e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2",[[973,4],[974,4],[975,4],[976,4],[977,4],[978,4],[979,4],[1129,4]]]
+[36,"/fixtures/0/stages/0",1,"19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e",[[2038,2],[2039,2],[2040,2],[2041,2],[2042,2],[2043,2],[2044,2],[2045,2],[2046,2],[2047,2],[2048,2],[2049,2],[2050,2],[2051,2],[2052,2],[2053,2],[2054,2],[2055,2],[2056,2],[2057,2],[2058,2],[2059,2],[2060,2],[2061,2],[2062,2]]]
+[36,"/fixtures/0/stages/1",2,"d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f",[[988,3],[992,3],[996,3],[1000,3],[1004,3],[1008,3],[1012,3],[1016,3],[1020,3],[1024,3],[1028,3],[1032,3],[1036,3],[1040,3],[1044,3],[1048,3],[1052,3],[1056,3],[1060,3],[1064,3],[1068,3],[1072,3],[1076,3],[1080,3],[1084,3]]]
+[36,"/fixtures/0/stages/2",3,"d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581",[[988,4],[992,4],[996,4],[1000,4],[1004,4],[1008,4],[1012,4],[1016,4],[1020,4],[1024,4],[1028,4],[1032,4],[1036,4],[1040,4],[1044,4],[1048,4],[1052,4],[1056,4],[1060,4],[1064,4],[1068,4],[1072,4],[1076,4],[1080,4],[1084,4]]]
+[36,"/fixtures/1/stages/0",2,"4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a",[[989,3],[993,3],[997,3],[1001,3],[1005,3],[1009,3],[1013,3],[1017,3],[1021,3],[1025,3],[1029,3],[1033,3],[1037,3],[1041,3],[1045,3],[1049,3],[1053,3],[1057,3],[1061,3],[1065,3],[1069,3],[1073,3],[1077,3],[1081,3],[1085,3]]]
+[36,"/fixtures/1/stages/1",3,"9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e",[[989,4],[993,4],[997,4],[1001,4],[1005,4],[1009,4],[1013,4],[1017,4],[1021,4],[1025,4],[1029,4],[1033,4],[1037,4],[1041,4],[1045,4],[1049,4],[1053,4],[1057,4],[1061,4],[1065,4],[1069,4],[1073,4],[1077,4],[1081,4],[1085,4]]]
+[36,"/fixtures/2/stages/0",2,"d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec",[[990,3],[994,3],[998,3],[1002,3],[1006,3],[1010,3],[1014,3],[1018,3],[1022,3],[1026,3],[1030,3],[1034,3],[1038,3],[1042,3],[1046,3],[1050,3],[1054,3],[1058,3],[1062,3],[1066,3],[1070,3],[1074,3],[1078,3],[1082,3],[1086,3]]]
+[36,"/fixtures/2/stages/1",3,"61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8",[[990,4],[994,4],[998,4],[1002,4],[1006,4],[1010,4],[1014,4],[1018,4],[1022,4],[1026,4],[1030,4],[1034,4],[1038,4],[1042,4],[1046,4],[1050,4],[1054,4],[1058,4],[1062,4],[1066,4],[1070,4],[1074,4],[1078,4],[1082,4],[1086,4]]]
+[36,"/fixtures/3/stages/0",2,"13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2",[[991,3],[995,3],[999,3],[1003,3],[1007,3],[1011,3],[1015,3],[1019,3],[1023,3],[1027,3],[1031,3],[1035,3],[1039,3],[1043,3],[1047,3],[1051,3],[1055,3],[1059,3],[1063,3],[1067,3],[1071,3],[1075,3],[1079,3],[1083,3],[1087,3]]]
+[36,"/fixtures/3/stages/1",3,"115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2",[[991,4],[995,4],[999,4],[1003,4],[1007,4],[1011,4],[1015,4],[1019,4],[1023,4],[1027,4],[1031,4],[1035,4],[1039,4],[1043,4],[1047,4],[1051,4],[1055,4],[1059,4],[1063,4],[1067,4],[1071,4],[1075,4],[1079,4],[1083,4],[1087,4]]]
+[37,"/fixtures/0/stages/0",2,"be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b",[[1088,3],[1089,3],[1090,3],[1091,3],[1092,3],[1093,3],[1094,3],[1095,3],[1096,3],[1097,3],[1098,3],[1099,3],[1100,3],[1101,3],[1102,3],[1103,3],[1104,3],[1105,3],[1106,3],[1107,3],[1108,3],[1109,3],[1110,3],[1111,3],[1112,3]]]
+[37,"/fixtures/0/stages/1",3,"8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37",[[1088,4],[1089,4],[1090,4],[1091,4],[1092,4],[1093,4],[1094,4],[1095,4],[1096,4],[1097,4],[1098,4],[1099,4],[1100,4],[1101,4],[1102,4],[1103,4],[1104,4],[1105,4],[1106,4],[1107,4],[1108,4],[1109,4],[1110,4],[1111,4],[1112,4]]]
+[38,"/records/0/stages/0",2,"7797f8b2f85799bd268bf06f71383c5e441791fb25d6c11afe368dde7b688f2a",[[965,3]]]
+[38,"/records/0/stages/1",3,"3ddf20d448ac238027732e5de25eceff5461797fae6dadac35f3ff3a32a85a89",[[965,4]]]
+[38,"/records/1/stages/0",2,"f1dcbdef6c39dbf4dd3542b151d9ca227d9ba45ff1e57d7f5d5ee7a93f09a0d4",[[967,3]]]
+[38,"/records/1/stages/1",3,"cc679e863a7f3653d5ef77a16942b4107959b879360a5b4cf0dd0c3d237cc3f6",[[967,4]]]
+[38,"/records/2/stages/0",2,"943b9548d42ec57747affb74e4de43cd2f8b6dffcc84b3f57c61c014622672e4",[[969,3]]]
+[38,"/records/2/stages/1",3,"46e7ba5d6bd8e53adf639c09a6f053a9d08edb608b10649a80c737ced39afaa4",[[969,4]]]
+[38,"/records/3/stages/0",2,"2029ebf4dd778b25d084af4e1383c8952caf0bd2f0c8e5940d2f82be7158e34d",[[970,3]]]
+[38,"/records/3/stages/1",3,"4b266bce7d640560951da1ff3eda5e0cc1550c8614b2a1ea6183519c8d722903",[[970,4]]]
+[38,"/records/4/stages/0",2,"468c19abe84ce7bca229d73c81d4ded6564c8d148b95f351a057e04b1b80098e",[[966,3]]]
+[38,"/records/4/stages/1",3,"8fb8eb364963656a48995d4052deca74cec8dfd07395d74788b843d7264a7128",[[966,4]]]
+[38,"/records/5/stages/0",2,"5696e1a6c73da853e281aa34b7d6d2fe6dc8be97c211b86c2896a53e6526045a",[[968,3]]]
+[38,"/records/5/stages/1",3,"75ac82bc1ef628d35241da2b78df5701adbe94efd1da970a2c40b47deb921515",[[968,4]]]
+[38,"/records/6/stages/0",2,"193a37484a4f7f25aa08e90c23783e2bbacb3cac45721e59fc60bd86d6b12c79",[[972,3]]]
+[38,"/records/6/stages/1",3,"b9455f8cdc76ec4612f947ce914f722c74032563c11075dad9caf80353c14623",[[972,4]]]
+[38,"/records/7/stages/0",2,"b8cf58c826726dd7c44c2c15f2530af1e33cbe393bd65e8de867ef14ed2276c3",[[971,3]]]
+[38,"/records/7/stages/1",3,"a19272331578b9876a758d37c2e72dcac32cc7151aa0b2aeb549388f74c8cb51",[[971,4]]]
+[39,"/records/0/stages/0",2,"34df0473cf3a71f266d58df0459e952a3e93d223f7635e3c0d81fdec210e8e0a",[[1113,3]]]
+[39,"/records/0/stages/1",3,"884148218b4dbe52217fe5aaceae1dbe2a6322f13e2861aba575e5041cdcb472",[[1113,4]]]
+[39,"/records/1/stages/0",2,"e12664cd3c2c8ab1ebee94128e90959dfd9fde36f4023cd59ad4ae6b317a1107",[[1115,3]]]
+[39,"/records/1/stages/1",3,"14dcc3ca9825b8567bc0ab8f60df13f8a003993172ec9d8b6463f31eb12b5b1e",[[1115,4]]]
+[39,"/records/2/stages/0",2,"1a94dd1d6973c3bba56fbbb5e2972c81d74f9afa874e0e6435de228453839aa4",[[1121,3]]]
+[39,"/records/2/stages/1",3,"b66dbbbbdb29f1a7d3e5cf07a95446ba1875fee261bb0b3e37206aa5c95c9e4d",[[1121,4]]]
+[39,"/records/3/stages/0",2,"206e31e80c020079ad30648d80a62afaf3564e99b9fc60bce10a6d013fe6c3a2",[[1122,3]]]
+[39,"/records/3/stages/1",3,"22fa426e26b6e51a27796d155fc1fe63e70a16bf4b36d73020d02c786769cd5e",[[1122,4]]]
+[39,"/records/4/stages/0",2,"5ef0b5eecdcc862411ed22bc193ddbcea8d94eb7d7f89158a7764a9759137fbc",[[1114,3]]]
+[39,"/records/4/stages/1",3,"782d840202c153255ac2ac63fdee9b5284b1fb992ec0e82e4fb388e11ba09780",[[1114,4]]]
+[39,"/records/5/stages/0",2,"940e92ff27f876255377efa918c901fed7e3e68e5ea77b987c1e267dc681f894",[[1116,3]]]
+[39,"/records/5/stages/1",3,"d4b51c1e1828a81d1cf6c7fe583ca851ab460871807158be6476b77382dedf9e",[[1116,4]]]
+[39,"/records/6/stages/0",2,"06528543fb56e5d6e216226520c60e3fba9c10caee88988d660dc3bf62cfc402",[[1124,3]]]
+[39,"/records/6/stages/1",3,"69d5cab5865f1616c640a859d6acfd3d704c8a4b6a0504edd360b513973809e3",[[1124,4]]]
+[39,"/records/7/stages/0",2,"f81cee09a2ed0f8c2a26a84a608b4d8363fb89c6da62fbc53ef36e4669d56432",[[1123,3]]]
+[39,"/records/7/stages/1",3,"bb9ce5c9f7687237a762d8ad90791c121859cc83620ca6ef2755a0ead7e93138",[[1123,4]]]
+[40,"/records/0/stages/0",2,"978ee747ff8cc5c08d711a0c53853888946906e2669cc24a21928ad8adf71e8e",[[1117,3]]]
+[40,"/records/0/stages/1",3,"018e9891184767f6e465cb1f24794c6c0145ef5077c2beeca95e42e05aa4189d",[[1117,4]]]
+[40,"/records/1/stages/0",2,"b74c85eca5faf76b32dcda134941e5b09e74cf1aadbfca395a09ad0a4f66286a",[[1119,3]]]
+[40,"/records/1/stages/1",3,"041f6695d70e0ae239ca5a008247c1c84c651ecdff45cb9a578559f6c0a27731",[[1119,4]]]
+[40,"/records/2/stages/0",2,"4aa071df474c2e447be1941f74662bbb1420c01e4fdab1c071aca9d571863b6f",[[1125,3]]]
+[40,"/records/2/stages/1",3,"ca07c7e22bb41deec0862f637e40df38b910149068f4abc8a1e480493439e7f3",[[1125,4]]]
+[40,"/records/3/stages/0",2,"72e7739e891781c3475cf8cacdf67a739e6623660d0f83794818612cd297a82e",[[1126,3]]]
+[40,"/records/3/stages/1",3,"a7361b01c024ab1c79cbe8744f497eaeba5eacbe7ead475ba79e21b317abe01f",[[1126,4]]]
+[40,"/records/4/stages/0",2,"2ae77445ff0a92724257c8a3fb55f6db72266289f6b97e6b47d21516466e5821",[[1118,3]]]
+[40,"/records/4/stages/1",3,"e6527d661a654f0d5d7b508d529f9008ea0aa02aca95bc47712d9669b74a4988",[[1118,4]]]
+[40,"/records/5/stages/0",2,"8c147cb0046175cd363c824b78ace3b1acdd06bb5c9d49990db97a715a2ef530",[[1120,3]]]
+[40,"/records/5/stages/1",3,"3859169c6918f08e624d5ace5c0745797f3757d9f96cad14b37655e71e37e405",[[1120,4]]]
+[40,"/records/6/stages/0",2,"b3d041020ef9ee6fb6013cc491495ec4add4a035da239b9d560b1017e0daeda7",[[1128,3]]]
+[40,"/records/6/stages/1",3,"c2b1813cc3f516d79a19fa139c7fc5d119fc1392db6b1f299fb6ef0ef46df6ca",[[1128,4]]]
+[40,"/records/7/stages/0",2,"309f99b05e8c94a7d67043662d247b6d2181ace2fa180f341d891848fe49df87",[[1127,3]]]
+[40,"/records/7/stages/1",3,"1677a6dc99df6612f9de3109d52d45b0293847849900d5ed579f651df96031ea",[[1127,4]]]
+[41,"/records/0/stages/0",2,"6fd14bda0a191ba64fe0a74811fd81acef22dae9a7364c814feb5dc6b82f25b6",[[980,3]]]
+[41,"/records/0/stages/1",3,"33a756465fa9b33d89f2b1ab0da1d5ce7a1f0e8293d97888b8db3e9cdc7a1f4a",[[980,4]]]
+[41,"/records/1/stages/0",2,"64edac3534d432072fb0cb9ee94ae758f539d41f98556515b3c8269967feb7a6",[[982,3]]]
+[41,"/records/1/stages/1",3,"9111691371ac05725c5828e7342d57b2c0dc17a189490d5fe06a4e499c657703",[[982,4]]]
+[41,"/records/2/stages/0",2,"cd7e55a90134cfc35729cd87003a44c454b151e8910822841d1d18f0c9ebc1a2",[[984,3]]]
+[41,"/records/2/stages/1",3,"c1566d75ad4047bc212b72c1d4e10c883b8616bb4276963654b1d5a2086c2aa8",[[984,4]]]
+[41,"/records/3/stages/0",2,"12003882bbbb7dab7f4bec48ce3a009d0b9d08e185f69fb277afceef476fb7c9",[[985,3]]]
+[41,"/records/3/stages/1",3,"e0029151f55c5f3974fcf5c52523d22289796d7bfbd8fa74916f3cee0a82459d",[[985,4]]]
+[41,"/records/4/stages/0",2,"0097b025872046cef28efa3c9f463bf8993bb9c41da22afc8477efc2a17a4ac3",[[981,3]]]
+[41,"/records/4/stages/1",3,"75f59d33a1fc63c903fefc73ee06a31dfa1364d3a5dedcaa5cda933152cc92d2",[[981,4]]]
+[41,"/records/5/stages/0",2,"b93a06888a9596b56f9ec7dec0e30eb7a87767aa6c4bab12aac88b0c80e4c7e0",[[983,3]]]
+[41,"/records/5/stages/1",3,"6fb8d4d5c0e41f500f6c523bd132db310bb25628a68ae9cb1331f0a890017d87",[[983,4]]]
+[41,"/records/6/stages/0",2,"c30b514d66dd4ef7d3898ed560a1144be98a172ef526ce4388beaf1f7c7dbd44",[[987,3]]]
+[41,"/records/6/stages/1",3,"766d57b5690df60d31209316dc540cfc2f8c31c303b3b3fae6d523385f764f40",[[987,4]]]
+[41,"/records/7/stages/0",2,"c10be1e6711b2a58cbe3fa9cf60de15556b9496ba056edee6036a3c20b8dd2d6",[[986,3]]]
+[41,"/records/7/stages/1",3,"5c07f0d6c2eea06bdaa939cf307a9323910485971dd341374537c4067524534c",[[986,4]]]
+[42,"/records/0/stages/0",2,"ae995db003010942ebdf8a541fb71e7a209b14d2035dc9b39ceebb44a20a8b22",[[949,3]]]
+[42,"/records/0/stages/1",3,"d3e0072eed343ab18506217d990ecc273e425433b2c7866bbf1cb84c84c48651",[[949,4]]]
+[42,"/records/1/stages/0",2,"512623d519f30b4de0fca57417959c5a3e4f6208ac12e3691744f3fbec21cefa",[[951,3]]]
+[42,"/records/1/stages/1",3,"0e549ea726768ab22519dacc89c24c3546e6cfcf709f8b53b5a8e86030aa8deb",[[951,4]]]
+[42,"/records/2/stages/0",2,"48f8afefe6dcc736f9b1e6750a9aa96bcbe0f1406d48a768f5ec04527b9f8ff2",[[957,3]]]
+[42,"/records/2/stages/1",3,"e41e8f8421b3066da06e8b9f77e52d19c248e2833be81e57cbe83b1dc4e3b34a",[[957,4]]]
+[42,"/records/3/stages/0",2,"cfc39a65a680d1cbbc4c17da9f23149ff569b977c427d1b56d9e22ae71752d6d",[[958,3]]]
+[42,"/records/3/stages/1",3,"ed9c523935c6e4d65c559d3b3246acf09a9515e0fade43ca29a79225e3009374",[[958,4]]]
+[42,"/records/4/stages/0",2,"0d1ce7e2d45fa78ac932971fa7ccc0f46e00d454de87b3e1ea926b0b22636b9b",[[950,3]]]
+[42,"/records/4/stages/1",3,"f8e6cb0ffef1b7c51d3caa4e5f589992e341d54ef3c6b0f0d1e8ac2581ae5db2",[[950,4]]]
+[42,"/records/5/stages/0",2,"402335a93decd5671794a0c5ec5e6f09ffd69ed6b26754fac45e7f294b3aac25",[[952,3]]]
+[42,"/records/5/stages/1",3,"11e848685fa3b253ee0b3abb5fee26cf6fb7b3e7f567351483da94bd54ed1afc",[[952,4]]]
+[42,"/records/6/stages/0",2,"7d2ab75eca2756a0c4d2c5544fe7837b92630aa42748c757bf68e0dcf6cc0c8f",[[960,3]]]
+[42,"/records/6/stages/1",3,"dea3c30a40b3bf4e28518ac19ce412aed32cb5f0394d48bb75ee09953527c0f4",[[960,4]]]
+[42,"/records/7/stages/0",2,"c8888ee61f060e895057d138bf3de9ec113eb39bb2c874055eda73abac1316e9",[[959,3]]]
+[42,"/records/7/stages/1",3,"770bae1810b48f7e160fb5a12fc8c49e4075d0c9cf30402bfa707c38346aa36a",[[959,4]]]
+[42,"/records/8/stages/0",2,"f178dc6498dfe1335f303815c77a74f6c64eef9102c2ae74312466d8ac126ab0",[[929,3]]]
+[42,"/records/8/stages/1",3,"e32824932666209d5959192f45f412c7f087fd40cd3b8a97e82e9004721855c5",[[929,4]]]
+[42,"/records/9/stages/0",2,"751f4ccde1a953061a6de9f03fc03e29bcab215d3e0dfaf5cadec2fdab2ffda9",[[931,3]]]
+[42,"/records/9/stages/1",3,"d4b7f6a5ffd37de7a86d2c35f122c69356b19e3ab05c2eaf64e726dd677fffd2",[[931,4]]]
+[42,"/records/10/stages/0",2,"aa743b0396151d7184adfa37f3ff722cd693b7bdb74f433a522498874947fbfb",[[933,3]]]
+[42,"/records/10/stages/1",3,"56253d8e8cd56f03dadc8369add61af230eac577a3258c42a28dcf4f2ede4f3d",[[933,4]]]
+[42,"/records/11/stages/0",2,"6c8d0ebc3993fe3649feb8dc574ca4aa6d9bbb8a54b57bab64439c72b3183f91",[[934,3]]]
+[42,"/records/11/stages/1",3,"e93fae493c1bff6c15d4b462fe8d3e9926f99da8c339f07247d5633515195e36",[[934,4]]]
+[42,"/records/12/stages/0",2,"f05e41f54c80df9399f1535e4630d9ac644c42c0c8bb66c19902effd9fe96f3b",[[930,3]]]
+[42,"/records/12/stages/1",3,"2733247c94953620e2bdfdda7e22bb9f6a0f731285caa41a50d70a50d63c92f3",[[930,4]]]
+[42,"/records/13/stages/0",2,"9e91aa50a469e4ae22cabf614f0c86c11a381dd99b7d57a4f376c7f86b815819",[[932,3]]]
+[42,"/records/13/stages/1",3,"c0e9109eafe381fc83456148a9518994753df9868388321037f0cbe11c64a260",[[932,4]]]
+[42,"/records/14/stages/0",2,"f13fef29cc4b67af7e5a181bef0b806732d7eea0ec0f5176e91581b982f20511",[[936,3]]]
+[42,"/records/14/stages/1",3,"8c0557fe188cb6b3e8cf492411a8e5ae3072d44fd5ab349e21951c62b0ed30a9",[[936,4]]]
+[42,"/records/15/stages/0",2,"75d63c6f58dbf64b7ff7c0bff1b89fd0c2c7776ecc86997ccfcd3184130f610f",[[935,3]]]
+[42,"/records/15/stages/1",3,"e3f4ba91d089260787eb84c1e93a4b9ad6ea30660ea314e755f025574e205657",[[935,4]]]
+[43,"/fixtures/3/stages/0",1,"d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385",[[64,2],[66,2],[242,2],[244,2],[1153,2],[1155,2]]]
+[43,"/fixtures/3/stages/1",2,"fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb",[[64,3],[66,3],[242,3],[244,3],[1153,3],[1155,3]]]
+[43,"/fixtures/3/stages/2",3,"bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6",[[65,4],[67,4],[243,4],[245,4],[1154,4],[1156,4]]]
+[43,"/fixtures/4/stages/0",1,"397d8179a2161540732e23d2358db3f5b9d44664430b730aac91c4dc83e5cded",[[295,2]]]
+[43,"/fixtures/4/stages/1",2,"7610b182b5f89de1c960aeb1cac1a4aff14a652935f535dc6e90da6fb8a211e8",[[295,3]]]
+[43,"/fixtures/4/stages/2",3,"5001111b2c815febac9426aed7a4b9a51ab752be9b2d97985f60a745e49cd0a7",[[296,4]]]
+[43,"/fixtures/0/stages/0",2,"9ec007869a74a8d4d9bf7be114f276c2c4e26f3e4c34eeaab721362bbf9a1918",[[1238,3],[1694,3]]]
+[43,"/fixtures/0/stages/1",3,"14f47ee4cc2165c26a56b268fb647e325de01ae94e284257de734b97dea0626a",[[1239,4],[1695,4]]]
+[43,"/fixtures/2/stages/0",2,"fb0578258494bd20b8dc20c6fe2243954b382980ac39ab5577a752f9af9a20e5",[[302,3]]]
+[43,"/fixtures/2/stages/1",3,"0236c2ff2b8a9c74dc5ef689d53ae080fbc93f5d23cee0b18105663b14080f9c",[[303,4]]]
+[43,"/fixtures/1/stages/0",2,"24cc1707db68cd33a711956ac50c8c65ab3e84eb5cf71a00fc15160d770b645d",[[239,3],[1689,3]]]
+[43,"/fixtures/1/stages/1",3,"7feb7ce1442470cb7ca3f613338c9fa901e16b23bc621a84ac3aa46a8583088d",[[240,4],[1690,4]]]
+[44,"/fixtures/0/stages/0",1,"b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016",[[1254,2],[1255,2],[1256,2],[1257,2]]]
+[44,"/fixtures/0/stages/1",2,"178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764",[[1254,3],[1255,3],[1256,3],[1257,3]]]
+[44,"/fixtures/0/stages/2",3,"efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf",[[1254,4],[1255,4],[1256,4],[1257,4]]]
+[45,"/fixtures/0/stages/0",2,"b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4",[[1272,3],[1592,3],[1593,3]]]
+[45,"/fixtures/0/stages/2",3,"0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f",[[1272,4],[1592,4],[1593,4]]]
+[46,"/fixtures/0/stages/0",2,"7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449",[[1276,3],[1277,3],[1278,3],[1279,3],[1280,3],[1281,3],[1282,3],[1283,3],[1284,3],[1285,3],[1286,3],[1287,3],[1288,3],[1289,3],[1290,3],[1291,3],[1292,3],[1293,3],[1294,3],[1295,3],[1296,3],[1297,3],[1298,3],[1299,3],[1300,3],[1301,3],[1302,3],[1303,3],[1304,3],[1305,3],[1306,3],[1307,3],[1308,3],[1309,3]]]
+[46,"/fixtures/0/stages/1",3,"6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34",[[1276,4],[1277,4],[1278,4],[1279,4],[1280,4],[1281,4],[1282,4],[1283,4],[1284,4],[1285,4],[1286,4],[1287,4],[1288,4],[1289,4],[1290,4],[1291,4],[1292,4],[1293,4],[1294,4],[1295,4],[1296,4],[1297,4],[1298,4],[1299,4],[1300,4],[1301,4],[1302,4],[1303,4],[1304,4],[1305,4],[1306,4],[1307,4],[1308,4],[1309,4]]]
+[47,"/fixtures/0/stages/0",2,"000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58",[[1316,3],[1317,3],[1318,3],[1319,3],[1320,3],[1321,3],[1322,3],[1323,3],[1324,3],[1325,3],[1326,3],[1327,3],[1328,3],[1329,3],[1331,3],[1332,3],[1333,3],[1334,3],[1335,3],[1336,3],[1337,3],[1338,3],[1339,3],[1340,3],[1341,3],[1343,3],[1360,3],[1361,3],[1362,3],[1363,3],[1364,3],[1365,3],[1366,3],[1367,3],[1368,3],[1369,3],[1370,3],[1371,3],[1372,3],[1373,3],[1374,3],[1375,3],[1376,3],[1377,3],[1378,3],[1379,3],[1380,3],[1381,3],[1382,3],[1383,3],[1384,3],[1385,3],[1386,3],[1387,3],[1388,3],[1389,3],[1390,3],[1391,3]]]
+[47,"/fixtures/0/stages/2",3,"6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499",[[1316,4],[1317,4],[1318,4],[1319,4],[1320,4],[1321,4],[1322,4],[1323,4],[1324,4],[1325,4],[1326,4],[1327,4],[1328,4],[1329,4],[1331,4],[1332,4],[1333,4],[1334,4],[1335,4],[1336,4],[1337,4],[1338,4],[1339,4],[1340,4],[1341,4],[1343,4],[1360,4],[1361,4],[1362,4],[1363,4],[1364,4],[1365,4],[1366,4],[1367,4],[1368,4],[1369,4],[1370,4],[1371,4],[1372,4],[1373,4],[1374,4],[1375,4],[1376,4],[1377,4],[1378,4],[1379,4],[1380,4],[1381,4],[1382,4],[1383,4],[1384,4],[1385,4],[1386,4],[1387,4],[1388,4],[1389,4],[1390,4],[1391,4]]]
+[48,"/fixtures/0/stages/0",2,"69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6",[[1344,3],[1345,3],[1346,3],[1347,3],[1348,3],[1349,3],[1350,3],[1351,3],[1352,3],[1353,3],[1354,3],[1355,3],[1356,3],[1357,3],[1358,3],[1359,3],[1620,3],[1621,3],[1622,3],[1623,3],[1624,3],[1625,3],[1626,3],[1627,3],[1628,3],[1629,3],[1630,3],[1631,3],[1632,3],[1633,3],[1634,3],[1635,3]]]
+[48,"/fixtures/0/stages/2",3,"e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2",[[1344,4],[1345,4],[1346,4],[1347,4],[1348,4],[1349,4],[1350,4],[1351,4],[1352,4],[1353,4],[1354,4],[1355,4],[1356,4],[1357,4],[1358,4],[1359,4],[1620,4],[1621,4],[1622,4],[1623,4],[1624,4],[1625,4],[1626,4],[1627,4],[1628,4],[1629,4],[1630,4],[1631,4],[1632,4],[1633,4],[1634,4],[1635,4]]]
+[49,"/fixtures/0/stages/0",2,"c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad",[[1393,3],[1401,3],[1409,3],[1417,3],[1425,3],[1435,3],[1443,3],[1451,3],[1459,3],[1467,3],[1475,3],[1484,3],[1492,3],[1500,3],[1508,3],[1516,3],[1525,3],[1533,3],[1541,3],[1549,3],[1557,3],[1565,3],[1573,3],[1581,3]]]
+[49,"/fixtures/0/stages/1",2,"16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69",[[1394,3],[1402,3],[1410,3],[1418,3],[1426,3],[1436,3],[1444,3],[1452,3],[1460,3],[1468,3],[1476,3],[1485,3],[1493,3],[1501,3],[1509,3],[1517,3],[1526,3],[1534,3],[1542,3],[1550,3],[1558,3],[1566,3],[1574,3],[1582,3]]]
+[49,"/fixtures/0/stages/2",2,"58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea",[[1395,3],[1403,3],[1411,3],[1419,3],[1427,3],[1437,3],[1445,3],[1453,3],[1461,3],[1469,3],[1477,3],[1486,3],[1494,3],[1502,3],[1510,3],[1518,3],[1527,3],[1535,3],[1543,3],[1551,3],[1559,3],[1567,3],[1575,3],[1583,3]]]
+[49,"/fixtures/0/stages/3",2,"91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c",[[1396,3],[1404,3],[1412,3],[1420,3],[1428,3],[1438,3],[1446,3],[1454,3],[1462,3],[1470,3],[1478,3],[1487,3],[1495,3],[1503,3],[1511,3],[1519,3],[1528,3],[1536,3],[1544,3],[1552,3],[1560,3],[1568,3],[1576,3],[1584,3]]]
+[49,"/fixtures/0/stages/4",2,"7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323",[[1397,3],[1405,3],[1413,3],[1421,3],[1429,3],[1439,3],[1447,3],[1455,3],[1463,3],[1471,3],[1479,3],[1488,3],[1496,3],[1504,3],[1512,3],[1520,3],[1529,3],[1537,3],[1545,3],[1553,3],[1561,3],[1569,3],[1577,3],[1585,3]]]
+[49,"/fixtures/0/stages/5",2,"c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51",[[1398,3],[1406,3],[1414,3],[1422,3],[1430,3],[1440,3],[1448,3],[1456,3],[1464,3],[1472,3],[1480,3],[1489,3],[1497,3],[1505,3],[1513,3],[1521,3],[1530,3],[1538,3],[1546,3],[1554,3],[1562,3],[1570,3],[1578,3],[1586,3]]]
+[49,"/fixtures/0/stages/6",2,"ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6",[[1399,3],[1407,3],[1415,3],[1423,3],[1431,3],[1441,3],[1449,3],[1457,3],[1465,3],[1473,3],[1481,3],[1490,3],[1498,3],[1506,3],[1514,3],[1522,3],[1531,3],[1539,3],[1547,3],[1555,3],[1563,3],[1571,3],[1579,3],[1587,3]]]
+[49,"/fixtures/0/stages/7",2,"0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1",[[1400,3],[1408,3],[1416,3],[1424,3],[1432,3],[1442,3],[1450,3],[1458,3],[1466,3],[1474,3],[1482,3],[1491,3],[1499,3],[1507,3],[1515,3],[1523,3],[1532,3],[1540,3],[1548,3],[1556,3],[1564,3],[1572,3],[1580,3],[1588,3]]]
+[49,"/fixtures/0/stages/11",3,"2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9",[[1393,4],[1401,4],[1409,4],[1417,4],[1425,4],[1435,4],[1443,4],[1451,4],[1459,4],[1467,4],[1475,4],[1484,4],[1492,4],[1500,4],[1508,4],[1516,4],[1525,4],[1533,4],[1541,4],[1549,4],[1557,4],[1565,4],[1573,4],[1581,4]]]
+[49,"/fixtures/0/stages/12",3,"83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917",[[1394,4],[1402,4],[1410,4],[1418,4],[1426,4],[1436,4],[1444,4],[1452,4],[1460,4],[1468,4],[1476,4],[1485,4],[1493,4],[1501,4],[1509,4],[1517,4],[1526,4],[1534,4],[1542,4],[1550,4],[1558,4],[1566,4],[1574,4],[1582,4]]]
+[49,"/fixtures/0/stages/13",3,"5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3",[[1397,4],[1405,4],[1413,4],[1421,4],[1429,4],[1439,4],[1447,4],[1455,4],[1463,4],[1471,4],[1479,4],[1488,4],[1496,4],[1504,4],[1512,4],[1520,4],[1529,4],[1537,4],[1545,4],[1553,4],[1561,4],[1569,4],[1577,4],[1585,4]]]
+[49,"/fixtures/0/stages/14",3,"ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87",[[1398,4],[1406,4],[1414,4],[1422,4],[1430,4],[1440,4],[1448,4],[1456,4],[1464,4],[1472,4],[1480,4],[1489,4],[1497,4],[1505,4],[1513,4],[1521,4],[1530,4],[1538,4],[1546,4],[1554,4],[1562,4],[1570,4],[1578,4],[1586,4]]]
+[49,"/fixtures/0/stages/15",3,"ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34",[[1399,4],[1407,4],[1415,4],[1423,4],[1431,4],[1441,4],[1449,4],[1457,4],[1465,4],[1473,4],[1481,4],[1490,4],[1498,4],[1506,4],[1514,4],[1522,4],[1531,4],[1539,4],[1547,4],[1555,4],[1563,4],[1571,4],[1579,4],[1587,4]]]
+[49,"/fixtures/0/stages/16",3,"4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395",[[1400,4],[1408,4],[1416,4],[1424,4],[1432,4],[1442,4],[1450,4],[1458,4],[1466,4],[1474,4],[1482,4],[1491,4],[1499,4],[1507,4],[1515,4],[1523,4],[1532,4],[1540,4],[1548,4],[1556,4],[1564,4],[1572,4],[1580,4],[1588,4]]]
+[50,"/fixtures/0/stages/0",2,"80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373",[[1594,3],[1595,3],[1596,3],[1597,3],[1598,3],[1599,3],[1600,3],[1601,3],[1602,3],[1603,3],[1604,3],[1605,3],[1606,3],[1607,3],[1608,3],[1609,3],[1610,3],[1611,3],[1612,3],[1613,3],[1614,3],[1615,3],[1616,3],[1617,3],[1618,3],[1619,3],[1636,3],[1637,3],[1638,3],[1639,3],[1640,3],[1641,3],[1642,3],[1643,3],[1644,3],[1645,3],[1646,3],[1647,3],[1648,3],[1649,3],[1650,3],[1651,3],[1652,3],[1653,3],[1654,3],[1655,3],[1656,3],[1657,3],[1658,3],[1659,3],[1660,3],[1661,3],[1662,3],[1663,3],[1664,3],[1665,3],[1666,3],[1667,3]]]
+[50,"/fixtures/0/stages/2",3,"5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27",[[1594,4],[1595,4],[1596,4],[1597,4],[1598,4],[1599,4],[1600,4],[1601,4],[1602,4],[1603,4],[1604,4],[1605,4],[1606,4],[1607,4],[1608,4],[1609,4],[1610,4],[1611,4],[1612,4],[1613,4],[1614,4],[1615,4],[1616,4],[1617,4],[1618,4],[1619,4],[1636,4],[1637,4],[1638,4],[1639,4],[1640,4],[1641,4],[1642,4],[1643,4],[1644,4],[1645,4],[1646,4],[1647,4],[1648,4],[1649,4],[1650,4],[1651,4],[1652,4],[1653,4],[1654,4],[1655,4],[1656,4],[1657,4],[1658,4],[1659,4],[1660,4],[1661,4],[1662,4],[1663,4],[1664,4],[1665,4],[1666,4],[1667,4]]]
+[51,"/fixtures/0/stages/0",2,"5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6",[[1268,3],[1269,3],[1270,3],[1271,3],[1273,3],[1274,3],[1275,3],[1310,3],[1311,3],[1312,3],[1313,3],[1314,3],[1315,3],[1330,3],[1342,3],[1392,3],[1433,3],[1434,3],[1483,3],[1524,3],[1589,3],[1590,3],[1591,3],[1668,3]]]
+[51,"/fixtures/0/stages/1",3,"3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f",[[1268,4],[1269,4],[1270,4],[1271,4],[1273,4],[1274,4],[1275,4],[1310,4],[1311,4],[1312,4],[1313,4],[1314,4],[1315,4],[1330,4],[1342,4],[1392,4],[1433,4],[1434,4],[1483,4],[1524,4],[1589,4],[1590,4],[1591,4],[1668,4]]]
+[52,"/fixtures/0/stages/0",1,"cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a",[[1683,2],[1685,2],[1687,2]]]
+[52,"/fixtures/0/stages/1",2,"6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a",[[1683,3],[1685,3],[1687,3]]]
+[52,"/fixtures/0/stages/2",3,"05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b",[[1684,4],[1686,4],[1688,4]]]
+[53,"/fixtures/0/stages/0",1,"445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714",[[85,2],[86,2],[87,2],[88,2],[89,2],[90,2],[91,2],[92,2],[93,2],[94,2],[1162,2]]]
+[53,"/fixtures/0/stages/1",2,"44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466",[[85,3],[86,3],[87,3],[88,3],[89,3],[90,3],[91,3],[92,3],[93,3],[94,3],[1162,3]]]
+[53,"/fixtures/0/stages/2",3,"e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f",[[85,4],[86,4],[87,4],[88,4],[89,4],[90,4],[91,4],[92,4],[93,4],[94,4],[1162,4]]]
+[53,"/fixtures/1/stages/0",1,"fcf9f4e3b994a7ccfade3699fa2c7393855b07aee83a70d3c449d4f4260098c4",[[95,2],[96,2]]]
+[53,"/fixtures/1/stages/1",2,"701c7338ba48b42868907f3900d1074b019fab8e257f5676f9a92e368c4718f1",[[95,3],[96,3]]]
+[53,"/fixtures/1/stages/2",3,"0635d30976c9afa0d905c53f55599ce0fa36a009828f00a5ae0f5283857a21de",[[95,4],[96,4]]]
+[53,"/fixtures/3/stages/0",2,"c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc",[[1669,3],[1670,3],[1671,3],[1672,3],[1673,3],[1674,3],[1675,3],[1676,3],[1677,3],[1679,3]]]
+[53,"/fixtures/3/stages/1",3,"a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391",[[1669,4],[1670,4],[1671,4],[1672,4],[1673,4],[1674,4],[1675,4],[1676,4],[1677,4],[1679,4]]]
+[53,"/fixtures/4/stages/0",2,"02683e3a77ca0ae4f160645826b8e3da2b9da126376f94ce15265ad34e8ca282",[[1678,3]]]
+[53,"/fixtures/4/stages/1",3,"9d36f1191de4ceed2adb6d065c793c92620591a4fbd3bafe95571a43268f4ec3",[[1678,4]]]
+[53,"/fixtures/2/stages/0",1,"a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b",[[253,2],[254,2],[255,2],[256,2],[257,2],[258,2],[259,2],[260,2]]]
+[53,"/fixtures/2/stages/1",2,"95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b",[[253,3],[254,3],[255,3],[256,3],[257,3],[258,3],[259,3],[260,3]]]
+[53,"/fixtures/2/stages/2",3,"34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727",[[253,4],[254,4],[255,4],[256,4],[257,4],[258,4],[259,4],[260,4]]]
+[54,"/fixtures/0/stages/0",1,"5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66",[[109,2],[110,2],[111,2],[112,2],[113,2],[114,2],[115,2],[116,2],[117,2],[118,2],[119,2],[120,2],[121,2],[122,2],[123,2],[124,2],[125,2],[126,2],[127,2],[128,2],[129,2],[130,2],[131,2],[132,2],[133,2],[134,2],[135,2],[136,2],[137,2],[138,2],[139,2],[140,2],[141,2],[142,2],[143,2],[144,2],[145,2],[146,2],[147,2],[148,2],[149,2],[150,2],[151,2],[152,2],[153,2],[154,2],[155,2],[156,2],[157,2],[158,2],[159,2],[160,2],[161,2],[162,2],[163,2],[164,2],[165,2],[166,2],[167,2],[168,2],[169,2],[170,2],[171,2],[172,2]]]
+[54,"/fixtures/0/stages/1",2,"a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc",[[109,3],[110,3],[111,3],[112,3],[113,3],[114,3],[115,3],[116,3],[117,3],[118,3],[119,3],[120,3],[121,3],[122,3],[123,3],[124,3],[125,3],[126,3],[127,3],[128,3],[129,3],[130,3],[131,3],[132,3],[133,3],[134,3],[135,3],[136,3],[137,3],[138,3],[139,3],[140,3],[141,3],[142,3],[143,3],[144,3],[145,3],[146,3],[147,3],[148,3],[149,3],[150,3],[151,3],[152,3],[153,3],[154,3],[155,3],[156,3],[157,3],[158,3],[159,3],[160,3],[161,3],[162,3],[163,3],[164,3],[165,3],[166,3],[167,3],[168,3],[169,3],[170,3],[171,3],[172,3]]]
+[54,"/fixtures/0/stages/2",3,"8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9",[[109,4],[110,4],[111,4],[112,4],[113,4],[114,4],[115,4],[116,4],[117,4],[118,4],[119,4],[120,4],[121,4],[122,4],[123,4],[124,4],[125,4],[126,4],[127,4],[128,4],[129,4],[130,4],[131,4],[132,4],[133,4],[134,4],[135,4],[136,4],[137,4],[138,4],[139,4],[140,4],[141,4],[142,4],[143,4],[144,4],[145,4],[146,4],[147,4],[148,4],[149,4],[150,4],[151,4],[152,4],[153,4],[154,4],[155,4],[156,4],[157,4],[158,4],[159,4],[160,4],[161,4],[162,4],[163,4],[164,4],[165,4],[166,4],[167,4],[168,4],[169,4],[170,4],[171,4],[172,4]]]
+[55,"/fixtures/0/stages/0",1,"efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581",[[84,2],[97,2],[98,2],[99,2],[100,2],[102,2],[103,2],[104,2],[105,2],[106,2],[107,2],[108,2],[173,2],[174,2]]]
+[55,"/fixtures/0/stages/1",2,"418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a",[[84,3],[97,3],[98,3],[99,3],[100,3],[102,3],[103,3],[104,3],[105,3],[106,3],[107,3],[108,3],[173,3],[174,3]]]
+[55,"/fixtures/0/stages/2",3,"45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa",[[84,4],[97,4],[98,4],[99,4],[100,4],[102,4],[103,4],[104,4],[105,4],[106,4],[107,4],[108,4],[173,4],[174,4]]]
+[55,"/fixtures/1/stages/0",1,"065ffbad7cc3f4a868f3de9e7f86d3a4fb1b0ea7047e1ddcdd578192e1fe101c",[[101,2]]]
+[55,"/fixtures/1/stages/1",2,"c6f52db1ccb491246b76348fb43a73fc961823d087d80432a2cec08dc36439f1",[[101,3]]]
+[55,"/fixtures/1/stages/2",3,"eea1fea3e3abbce7a6d377fafd8ad8e2c8fec998a990afbd664b7ce4f3856dff",[[101,4]]]
+[56,"/fixtures/0/stages/0",1,"ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8",[[1696,2],[1697,2],[1698,2]]]
+[56,"/fixtures/0/stages/1",2,"ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3",[[1696,3],[1697,3],[1698,3]]]
+[56,"/fixtures/0/stages/2",3,"2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8",[[1696,4],[1697,4],[1698,4]]]
+[57,"/records/12/stages/0",1,"d23e19589de35127bff359facfbc56c4c16a19fd1aeef8a4568c6080ad4c1a68",[[430,2]]]
+[57,"/records/12/stages/1",2,"8eb444c30aad31f7ce8a70bcd06b91bae96ddd810e5c0e0bad67d669bba76463",[[430,3]]]
+[57,"/records/12/stages/2",3,"8a38549cf45825e7cf005005dcaa96cbdff09dd83f2b672a3d5bc23b8102018e",[[430,4]]]
+[57,"/records/13/stages/0",1,"078c3ec1a8767b2764c1a5dd2617806f672124e023a5c9fe5f3c40c6f5a36219",[[431,2]]]
+[57,"/records/13/stages/1",2,"9f448dfcfc8dc1e19e552c3f601db928004ab0f095c275cf846bc4e8adc3eceb",[[431,3]]]
+[57,"/records/13/stages/2",3,"5a759f49ba62f6028179716e660082e3b36877cefbe37f9ad1aa91c27aebc22c",[[431,4]]]
+[57,"/records/14/stages/0",1,"2228a4bf5e0bbaf7685dcbd5f0e890507df94d30aa9165cbe879037b84c74a4e",[[432,2]]]
+[57,"/records/14/stages/1",2,"c86c69e86d0084dd76afccfa7dd76416f5a906eda8f245fd5eaadfb94eeedfea",[[432,3]]]
+[57,"/records/14/stages/2",3,"b42051503a9dcfd9cab9b052388ae403c55c7b81a776697ba93d8146a265b42a",[[432,4]]]
+[57,"/records/15/stages/0",1,"a4b6c14f6dd0fdb52ad2e22873b19609ff47a1c8994bc5e3ce39c88bdc59a58f",[[433,2]]]
+[57,"/records/15/stages/1",2,"51c2def02e859676a80a71edbbd6563d7dac75a59fb25899f3718a20705b267d",[[433,3]]]
+[57,"/records/15/stages/2",3,"67149c3d0b2f90b47e4e1f3f361f390adea9ffd4186e04bafaabd93b8091168f",[[433,4]]]
+[57,"/records/16/stages/0",1,"28e76c919b457758f8eeaf5f976e581db31431ae31ed1efbc30a050a31da1344",[[435,2]]]
+[57,"/records/16/stages/1",2,"c74dd5efb5771ae005406a00b33baa93333dd3fbcaa2787d0dee90957c48871f",[[435,3]]]
+[57,"/records/16/stages/2",3,"f5dbfbc0015f15b42f13d4bb3e15a049ffe769e096166c94bbb0aecfaa567d43",[[435,4]]]
+[57,"/records/17/stages/0",1,"585528599fab5e664c529894e42a2b9604bc6e503f9eefc9d2117235a0fd35fd",[[434,2]]]
+[57,"/records/17/stages/1",2,"3d0cc72dd11e3399dbfe72a912f451f36835e6c60ff54db70e1b6589b905dbbc",[[434,3]]]
+[57,"/records/17/stages/2",3,"0b8c31b7c8cbef8fc5d164d732bfa611edc166ee875735791b5446e77f385f2f",[[434,4]]]
+[57,"/records/18/stages/0",2,"7769e680b674aa15719338a21a264d9e6a8bd0e7f4a3f90192b4f5507f2472cd",[[1158,3]]]
+[57,"/records/18/stages/1",3,"3a37572911b7e3d97e3785be11b99068c4b530eac6b0a16d6d74d6a0c6a56702",[[1159,4]]]
+[57,"/records/19/stages/0",2,"e2537a05bf31f004570ad178a6290d201b70855af40a531e7bb3c4930907b72f",[[1157,3]]]
+[57,"/records/19/stages/1",3,"5d3e002fed008a798ac7c23848acfa14f680ffa01f2d4c11154a6b127f3b71ce",[[1157,4]]]
+[57,"/records/29/stages/0",1,"de5ba8b2a4efbc2ca94148badabcdf0d50913a97a1663fb623592433f484e88e",[[234,2]]]
+[57,"/records/29/stages/1",2,"9c42148d5c5a275ca9ece93315e98eb0e9508a0b4e8cf0eba8cc761268f29c42",[[234,3]]]
+[57,"/records/29/stages/2",3,"d9b7b74a68441f7f40d480ab430166c946704ea11caf1a7e9951b4726130bc65",[[235,4]]]
+[57,"/records/30/stages/0",1,"d4565681a3a6f6f64ea89b63e96c44c1aef2d594b01f3c695eaa4d35147c250f",[[236,2]]]
+[57,"/records/30/stages/1",2,"eec368e7d9686a4072001c8362bc69ea91a9b25fb950b50232244c96e72c0887",[[236,3]]]
+[57,"/records/30/stages/2",3,"159beb8496be094d32d2c6421e8993b91ee8e0d8b787fa5313af73ea5af66eac",[[237,4]]]
+[57,"/records/31/stages/0",1,"cf14ce2498510d2ccdf8e26062d9dab33a2f018ab44d7402f731ee721edd9586",[[230,2]]]
+[57,"/records/31/stages/1",2,"5004901d9a7fa9fc719218c051ffd4362bcef5a21c7bcca397e74fdb199f3733",[[230,3]]]
+[57,"/records/31/stages/2",3,"c5f55a84bfd467a639f8fe3020d3fa05a6616160125d05ecd44cf631c78a954c",[[231,4]]]
+[57,"/records/32/stages/0",1,"6bba537191d91e4c7f577889e7832f879421a5a8b021a98982af2c068ecd5ac5",[[232,2]]]
+[57,"/records/32/stages/1",2,"ad295ebd24c5fa94a1990f0e676049b7232bf0421f6419e2230f054510b87ede",[[232,3]]]
+[57,"/records/32/stages/2",3,"aef5395fd4ebfc37a0a4fb468bb18c83a363528cd307bfeb7ed1218c5fd84fc2",[[233,4]]]
+[57,"/records/33/stages/0",1,"25b383a09f3293d7f5ceef1bc66d8400702d7a5c68d73f24532c371030e69c11",[[1265,2]]]
+[57,"/records/33/stages/1",2,"26b139fc6199abb20c9a7d92ddb0a30839fb2bdd3eafe285ec63bb49da4bb3c5",[[1265,3]]]
+[57,"/records/33/stages/2",3,"e287209ed91d6c476b48a3d08399a0911d64f601f3ae07e328db3f975fc2d2e9",[[1266,4]]]
+[57,"/records/43/stages/0",1,"4a17f68d96e936c632dbeb06ea74eae28b193c05cdfdc08373f0715f348eacf1",[[3,2]]]
+[57,"/records/43/stages/1",2,"246910573ff59e4feb1a3b8880eb69d36c7a14f49debfe41f1207db7b4ea51d4",[[3,3]]]
+[57,"/records/48/stages/0",1,"d5b275af3bb7cfa86185e4ce735308f68edfb3dd4a71261c8ed80131f8f67b84",[[1263,2]]]
+[57,"/records/48/stages/1",2,"9e7bf824b7fcf96ef357d4279091d73c4cd95081c9a10994d3a2abc4df06f58d",[[1263,3]]]
+[57,"/records/48/stages/2",3,"395cf308dc13aa17943c5d073be46dc163c23cd9add8444b68a0dcff491240de",[[1264,4]]]
+[57,"/records/49/stages/0",1,"4c8015a293ea78a1394796bab968f8d4b065b433149894cf07283122ab75837d",[[1220,2]]]
+[57,"/records/49/stages/1",2,"dcc151e0a9e33ce7a059a418221504ddf54bb43d6f1b8927cffd7cdd76c93939",[[1220,3]]]
+[57,"/records/49/stages/2",3,"0dab85a67df2645eacc4979c404b7be148188a35e648c7e180b82aba4269ae71",[[1221,4]]]
+[57,"/records/50/stages/0",1,"cc205d40d08b4db407d73012b4e02bacca951cbf2407578220e1a9993241f7c4",[[1232,2]]]
+[57,"/records/50/stages/1",2,"6d1ab9e1df5d932129b785eb98a099581f503fadd7a7e22b4e7621f06dd4ecce",[[1232,3]]]
+[57,"/records/50/stages/2",3,"4f4465879f688e4239d1fc44f8e1c2008923c890cf59e0d1e2cad6653348622d",[[1233,4]]]
+[57,"/records/51/stages/0",1,"12f77b50bd9a2ac2e167e43862d4925890afbc944ce1be9abf7a42bef834a693",[[1214,2]]]
+[57,"/records/51/stages/1",2,"e868aa2aec22540ec973c73eda5ad85c82e432f562a4956b89fc5ec681ff12c4",[[1214,3]]]
+[57,"/records/51/stages/2",3,"56902b3d564c19bb5e6ffca02b1f0f73c0c6ea1a440d7ba209fe9242e86d6b50",[[1215,4]]]
+[57,"/records/52/stages/0",1,"d6fef706af3d2993713934e1f5156c47628a6c943899ad369edf140f7b9f5c57",[[1226,2]]]
+[57,"/records/52/stages/1",2,"7e9ac34403d0a172c008ea25d9e62a0786926818eec2efb2091811ddca6c0caf",[[1226,3]]]
+[57,"/records/52/stages/2",3,"faebc8c202740b059817f861bd3cded8188e653457a4b9791ce87506a1c63314",[[1227,4]]]
+[57,"/records/53/stages/0",1,"d782063e4b1ae122d58089206151ace8d5b7d4f660a5d5159fb1b770f56b46a0",[[1218,2]]]
+[57,"/records/53/stages/1",2,"e4775aad2f0dbe6b4f3fc2db7d15985eed28a8adcc2031100d3db5a4c25600a4",[[1218,3]]]
+[57,"/records/53/stages/2",3,"578376a2dccd0ed45a721f981ae2bf7094d65d90e46479f5569dc8a00354275f",[[1219,4]]]
+[57,"/records/54/stages/0",1,"fa450ce0f0bf68458c30deaa762636db4ffcfbd33d8f8c5d627a057b59740efb",[[1230,2]]]
+[57,"/records/54/stages/1",2,"f67d1796c2477484afa60ecc1be1bc640a045537ba52e1512f1cb17bc46898e4",[[1230,3]]]
+[57,"/records/54/stages/2",3,"4ab01d57f019c3f6994b977e6528bbadebb9db1980ff1e4e57480354c0cb1aa0",[[1231,4]]]
+[57,"/records/55/stages/0",1,"fe998b216b523ddba326d894f7fd6c1c811b7b8f6793fd67fa9f32bb5426fe93",[[1212,2]]]
+[57,"/records/55/stages/1",2,"e822e31752229f9d5f8379f58427407f4da67974fdf5cb800973a1d420d99843",[[1212,3]]]
+[57,"/records/55/stages/2",3,"78a64562571e8dfa2d02c779947163cd4c087bf0fb45cfc9b26b7a9448da1d34",[[1213,4]]]
+[57,"/records/56/stages/0",1,"5a97a689d300dd7c199c714c01189afb154122960104cfc1a1f1e1180c4e7cf3",[[1224,2]]]
+[57,"/records/56/stages/1",2,"a3ed93208637ba99cf7cdbdd4ddbcc7dd13f263fcecc0842c46753c8c841ca34",[[1224,3]]]
+[57,"/records/56/stages/2",3,"01942dd79db0ae31a8cb526a0bab0e1bbac0a00e9494f7fd2c1ac989af6a68b2",[[1225,4]]]
+[57,"/records/57/stages/0",2,"00eed2a3ce51403c06b4557ca323d691094cf59685823dd73b0b329a2fc40d9a",[[1222,3]]]
+[57,"/records/57/stages/1",3,"d1301d100cbe1985f544b748d01b5a92e32173c169be106074f0614f23f7a08d",[[1223,4]]]
+[57,"/records/58/stages/0",2,"a423fb03456e022c274ab0aee6f4480239603e12b7657f78a3cce23264b020e6",[[1234,3]]]
+[57,"/records/58/stages/1",3,"41a4b0b1e404dde5ac1a6f457e13bde66b3a26ce037c64f3c691fa4666f07d5d",[[1235,4]]]
+[57,"/records/59/stages/0",2,"8dc004e784b4ad1d4151d0fbbb8c7ab4815437672e35cb37c8408d685938d961",[[1216,3]]]
+[57,"/records/59/stages/1",3,"ac475b6fa0fb52beb3890ebaba00ef224041684fd47bfec82f609768fb0ad71f",[[1217,4]]]
+[57,"/records/60/stages/0",2,"a83c401208ac85d4c6fe20f942147f8bd8410442b4ce60b86079b138ceac2208",[[1228,3]]]
+[57,"/records/60/stages/1",3,"a85d3dccb30f9491380b1f54913e7261c58667a5c22f195298c6a4f0738f3992",[[1229,4]]]
+[57,"/records/61/stages/0",2,"f847702ab3d3db2051359e9ba8869735a57a5cf5b92a2901b154340dc14cc080",[[261,3]]]
+[57,"/records/61/stages/1",3,"aa10765bb36d761080b94f2c0788a77e387b12032c06504740e5ff756239cda5",[[261,4]]]
+[57,"/records/62/stages/0",2,"903e4771d5a308e61a37282dd1f236481315e6519d5c06dbfd19c6c47f97acfd",[[271,3]]]
+[57,"/records/62/stages/1",3,"adc9face773753523147665457351aa23a8b43da9e697a565109e15870383010",[[271,4]]]
+[57,"/records/63/stages/0",2,"2c103ab626814fb233a6e15e90a48c2a6a5b78f7f13e38c4c8c6cebca28ef0d0",[[4,3]]]
+[57,"/records/63/stages/1",3,"af1befea9f33a173ed1a34fa549dfd6e8cf85f450042ec4948cdcd9eef20051f",[[4,4]]]
+[57,"/records/64/stages/0",2,"379f54089c227f714a7a906db098eef11d0ac042db017cbd022236e2f366c440",[[1258,3]]]
+[57,"/records/64/stages/1",3,"7e912f52acc5af50229ee7d87a37888ffbfd4dddc236e582d317512c281864d0",[[1258,4]]]
+[57,"/records/65/stages/0",2,"f4e083050e3dc987284bceea5b0064f86a0c78bca3754d8a92284911bbd7fe78",[[1131,3]]]
+[57,"/records/65/stages/1",3,"c6f92a98210d4280598f8059312562b774c73bae8a2baf9af8252031d8f5ac5b",[[1131,4]]]
+[57,"/records/66/stages/0",2,"f59f875a1a686231cf0e0c09cd8ea1d7107aaf7ba547e54f0847ba4d71726b27",[[503,3]]]
+[57,"/records/66/stages/1",3,"51930c796a141b52e8382f91a54de007b9908945d7bdddf8085b3c9e4575557a",[[503,4]]]
+[57,"/records/67/stages/0",2,"1bcd79e15bcc5f861613af0657054804f2cc98de0bc44082c84eeb6e06750d11",[[444,3]]]
+[57,"/records/67/stages/1",3,"37e990ab010dfd7a6aa363c1bc12abfb5d4e121bceb7b89ac8056454e95a2385",[[444,4]]]
+[57,"/records/68/stages/0",2,"e2dca2ee0088dbadc0cb0904d95659a2c0d5e7974da0c5e8aff75f0e290504a1",[[1150,3]]]
+[57,"/records/68/stages/1",3,"779e29da03300e6d8024f6c98694b6bbad23c127181f299430b3a6b09c8541b5",[[1150,4]]]
+[57,"/records/69/stages/0",2,"fc97be133b128891fbdca9f6a4b9769cfd2712fa35ee81884e52b9b436f6ef32",[[0,3]]]
+[57,"/records/69/stages/1",3,"cbafd197892b5b499ef62752012ccb5e86318641dd9784942872a4ca7d685893",[[0,4]]]
+[57,"/records/70/stages/0",1,"e7b66943f3e72c942117544102ede74b7d46537dc9c8f09cb4c800b1a8150b25",[[1712,2]]]
+[57,"/records/70/stages/1",2,"260159c4d2bde9577b33387b20e93e5e1e00605c7ce69ce157c5c5046bbaabf0",[[1713,3]]]
+[57,"/records/70/stages/2",3,"160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c",[[1714,4]]]
+[57,"/records/70/stages/3",2,"625e5da377d73334b52ec4bdd6bc6702e15b196ca54350f0b070cb61e09b60ff",[[191,3]]]
+[57,"/records/70/stages/4",3,"160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c",[[191,4]]]
+[57,"/records/71/stages/0",2,"5e1cf777d5d76a7579b7c840ac1250cd7961888126aeef92b619e2bfcf5718aa",[[262,3]]]
+[57,"/records/71/stages/1",3,"f962661dab009881f23af0cdf33aed0bdb5ea13629f790897da04f271d8cc305",[[263,4]]]
+[57,"/records/72/stages/0",2,"67da91ea322e2d60af10838e8a20890d99c38e07262abe9a6848ea5e1fbbf2fb",[[272,3]]]
+[57,"/records/72/stages/1",3,"871b4ac65868041bcc0f028deaaac2793f5df9db4518fdbfaa03580e9c26414a",[[272,4]]]
+[57,"/records/73/stages/0",2,"3b7f49dda4585f12ddfc9ecaf71517120de1aefd78a9f843fcb511f81bbabaf4",[[5,3]]]
+[57,"/records/73/stages/1",3,"b33cc169935b856e4ff58245496b889d56c48c772653634f963facb395c24326",[[6,4]]]
+[57,"/records/74/stages/0",2,"260eb7ac7c16c07d7eb3ac88e498b18d9e9f37b1c763d3dca78075888ea7cce8",[[1259,3]]]
+[57,"/records/74/stages/1",3,"cdc2a8724f570ba9cf569ea9240cf60a937116228c42a5d0d9bdcef1af591d35",[[1260,4]]]
+[57,"/records/75/stages/0",2,"0642183f45157615161325bf3fd8c344f4d8efea483a8bf2c111fd9bd7c9dc4f",[[1132,3]]]
+[57,"/records/75/stages/1",3,"090bddd2dbb7c5459d46de7c6956de9084a097a5041086b13a2a85322673072d",[[1133,4]]]
+[57,"/records/76/stages/0",2,"6d54f22a9e49d8c8e53739fb3942a54749c50fe7749970cd67c4c77932fc1e11",[[505,3]]]
+[57,"/records/76/stages/1",3,"34e9c8ebf06b8ee9b1df5eda184d80fb12116914e533e5f11ecf26f337751abc",[[505,4]]]
+[57,"/records/77/stages/0",2,"a4070d164dfb7ec86737afcc51fdef8e9452193df799721221b988492eab3188",[[446,3]]]
+[57,"/records/77/stages/1",3,"f9dc936add8025c51ff86d52cb72929dbbd1197519c0ffed33b53ba5eae61c7c",[[446,4]]]
+[57,"/records/78/stages/0",2,"d0e44e1921be6e1f3bc1cc0e885622b5bb8a552876f0f300d6cff83701c2b0a8",[[1151,3]]]
+[57,"/records/78/stages/1",3,"a6abb77861c39cdcaed05cd8512d1b2ca13a32c045cecedb72aa4038fc9c0d6e",[[1152,4]]]
+[57,"/records/79/stages/0",2,"7282ee206f4fdb3a54d1fe47412d9ab41a11954358a8e0ee203f434988aa5191",[[1,3]]]
+[57,"/records/79/stages/1",3,"93adc93cea87ea659d8a0b4542f44e1d897cb4c1cd53a06dc08becdac7de7ea6",[[2,4]]]
+[57,"/records/80/stages/0",1,"23c48425228bb21090b442cbcb63b58874efaf6cc05dde2ad7549aff23e3f8ce",[[1709,2]]]
+[57,"/records/80/stages/1",2,"7c94821ab335a6f02f2b3ca52d1e0c2428d1d26aac5e7c62852ef35cbfcba8e4",[[1710,3]]]
+[57,"/records/80/stages/2",3,"8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d",[[1711,4]]]
+[57,"/records/80/stages/3",2,"009a800094f29371a5f120862eb39d45b097aaa5715ce2e7ac26a43ec3665773",[[190,3]]]
+[57,"/records/80/stages/4",3,"8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d",[[190,4]]]
+[57,"/records/81/stages/0",1,"2bccf7085ce3ccc76e323145e0cd1e86fda391d0411adc51c1eea4ce806a2f03",[[1748,2]]]
+[57,"/records/81/stages/1",2,"5790119795c6ad9cbfc37ff239e9a123b44c05bde5d38fc5675565933d23a6ff",[[1749,3]]]
+[57,"/records/81/stages/2",3,"342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e",[[1750,4]]]
+[57,"/records/81/stages/3",2,"6833e6724886aa52db6de2c812bc6e0e2a7216e33ba07501701559eb4c68cf5c",[[213,3]]]
+[57,"/records/81/stages/4",3,"342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e",[[213,4]]]
+[57,"/records/82/stages/0",1,"f4b2d11b6f47357e52b5da2bd8374d6043945e2d3c7d28f0d9bce741658cbbfb",[[1730,2]]]
+[57,"/records/82/stages/1",2,"a886266d7776914c367d748d72bd9b1032db99c5eb5366f902a3b5b8e63f2add",[[1731,3]]]
+[57,"/records/82/stages/2",3,"55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8",[[1732,4]]]
+[57,"/records/82/stages/3",2,"644949d91a98e2c1b36b125ca7211d181b486d46e6b0335c823f0f78de9860cc",[[197,3]]]
+[57,"/records/82/stages/4",3,"55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8",[[197,4]]]
+[57,"/records/83/stages/0",1,"e02262263ec043530dec4469b22cfd6e831bc597d97b59e20132167985ae8e75",[[1721,2]]]
+[57,"/records/83/stages/1",2,"c797710407ad5b5c6425fc3ae32587456573e2442fb1398a784e61ab025c7f6b",[[1722,3]]]
+[57,"/records/83/stages/2",3,"d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806",[[1723,4]]]
+[57,"/records/83/stages/3",2,"336ac20640c8b6451d754f6afed9479d89a22cd5c47b6606e57f251337b35e0c",[[194,3]]]
+[57,"/records/83/stages/4",3,"d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806",[[194,4]]]
+[57,"/records/84/stages/0",1,"bb6bcfe8642c6981459470ff9b577231bb7c5575846df5a6b524e3ed834bfc4c",[[1745,2]]]
+[57,"/records/84/stages/1",2,"9dcafda6629d6de7301da44bf1e7690a5d1e8131d595cc975ea4e8b68a61de5d",[[1746,3]]]
+[57,"/records/84/stages/2",3,"3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6",[[1747,4]]]
+[57,"/records/84/stages/3",2,"bd97b8bfbf98ce29b538e25417e266f1d5c29d1d18b0789af1dd439fe4431d90",[[212,3]]]
+[57,"/records/84/stages/4",3,"3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6",[[212,4]]]
+[57,"/records/85/stages/0",2,"32b430781d5a0e49a3cf431871f05da67c5f0a37e984050e1a2030c9f4e4f283",[[1691,3]]]
+[57,"/records/85/stages/1",3,"9b62e57d3cf50fcc7ced943d8501901809cc7d321cab2fbfb101791ed0909718",[[1692,4]]]
+[57,"/records/86/stages/0",2,"612fb1dc878abe6a363a52b221b9b188b38d5707df5567d98e8e0c26bd0902b4",[[56,3]]]
+[57,"/records/86/stages/1",3,"66e4eb480c47c5102813d6f0b276ec783bf14a6f3bdc849b69eca9c086bfed5c",[[57,4]]]
+[57,"/records/87/stages/0",2,"6af969aa9d4c327529700e8a82bc7e08ce798613e609091e74e47a57adaf113f",[[1160,3]]]
+[57,"/records/87/stages/1",3,"b7c77d308d6af079d03e08c447e632722a4dbb282c7d6faea32b7e1174e9f623",[[1161,4]]]
+[57,"/records/88/stages/0",2,"4d774d39425c57263a42ec917b671796b75c2e755a81255dfc72a40232e32367",[[1130,3]]]
+[57,"/records/88/stages/1",3,"85c6f2304703e7ee71bd8ff7bbb2dec8730ea51244c37934a0cc838b2d00ab69",[[1130,4]]]
+[57,"/records/128/stages/0",2,"80a7eb25c335cf849bc1e0be9ed1e391b9d311a1573d38be927d6aa35c8b4aa3",[[265,3]]]
+[57,"/records/128/stages/1",3,"83de32412117576acec7391b12d4c1e5f2f026f4fed9281689ec4a87e9b1deaf",[[266,4]]]
+[57,"/records/129/stages/0",2,"b7ccf304fffd63a678dec5f5775eb1fbdc60e0a295b7b2a9e878339cf8a0922a",[[293,3]]]
+[57,"/records/129/stages/1",3,"ad17848cb5677cf0a68e6e3eb435b41a165f6ef1cbaf69b0fa3e2b4d8bbfdcfe",[[294,4]]]
+[57,"/records/130/stages/0",2,"9bed831d23f91ca28a7fa178103ea679a4765ec587f918986874e52b21d00da1",[[269,3]]]
+[57,"/records/130/stages/1",3,"6a185d53a7cf04592fc205ee4ef0926f252fa57f5c50f1ee510c7aa6210578ee",[[270,4]]]
+[57,"/records/131/stages/0",2,"93c9c1a8aaa9ac2bee924f4b1eee264e0aa38c8c6f01c04bddf3bfd58d169bc4",[[268,3]]]
+[57,"/records/131/stages/1",3,"582782aeec6a6286313dd28330e3de2a23eeaa07c479ee6271c831e0c8f9d1e7",[[268,4]]]
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cp-async-mbarrier.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cp-async-mbarrier.json
deleted file mode 100644
index 47a162b936..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cp-async-mbarrier.json
+++ /dev/null
@@ -1,206 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "cp_async_mbarrier_arrive",
-      "source_record": "int_nvvm_cp_async_mbarrier_arrive",
-      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive",
-      "llvm_arguments": ["ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive(ptr) declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the generic-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed generic-address cp.async mbarrier arrival",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM bridge probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cp",
-        "modifiers": ["async", "mbarrier", "arrive", "b64"],
-        "operands": [{ "kind": "address" }]
-      }
-    },
-    {
-      "id": "cp_async_mbarrier_arrive_shared",
-      "source_record": "int_nvvm_cp_async_mbarrier_arrive_shared",
-      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive.shared",
-      "llvm_arguments": ["shared_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive.shared(ptr addrspace(3)) declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed shared-address cp.async mbarrier arrival",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.shared.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM bridge probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cp",
-        "modifiers": ["async", "mbarrier", "arrive", "shared", "b64"],
-        "operands": [{ "kind": "address" }]
-      }
-    },
-    {
-      "id": "cp_async_mbarrier_arrive_noinc",
-      "source_record": "int_nvvm_cp_async_mbarrier_arrive_noinc",
-      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive.noinc",
-      "llvm_arguments": ["ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive.noinc(ptr) declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the generic-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed generic-address no-increment cp.async mbarrier arrival",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.noinc.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM bridge probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cp",
-        "modifiers": ["async", "mbarrier", "arrive", "noinc", "b64"],
-        "operands": [{ "kind": "address" }]
-      }
-    },
-    {
-      "id": "cp_async_mbarrier_arrive_noinc_shared",
-      "source_record": "int_nvvm_cp_async_mbarrier_arrive_noinc_shared",
-      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive.noinc.shared",
-      "llvm_arguments": ["shared_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.cp.async.mbarrier.arrive.noinc.shared(ptr addrspace(3)) declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed shared-address no-increment cp.async mbarrier arrival",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.noinc.shared.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM bridge probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cp",
-        "modifiers": ["async", "mbarrier", "arrive", "noinc", "shared", "b64"],
-        "operands": [{ "kind": "address" }]
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cp-async.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cp-async.json
deleted file mode 100644
index 9e979406ae..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cp-async.json
+++ /dev/null
@@ -1,241 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "cp_async_ca_4",
-      "source_record": "int_nvvm_cp_async_ca_shared_global_4",
-      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.4",
-      "llvm_arguments": ["shared_ptr", "global_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.ca.shared.global.4 declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-all four-byte call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 4."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"4"}]}
-    },
-    {
-      "id": "cp_async_ca_zfill_4",
-      "source_record": "int_nvvm_cp_async_ca_shared_global_4_s",
-      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.4.s",
-      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.ca.shared.global.4.s declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-all four-byte source-size calls","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted both register and immediate source-size forms."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"4"},{"kind":"register_or_immediate"}]}
-    },
-    {
-      "id": "cp_async_ca_8",
-      "source_record": "int_nvvm_cp_async_ca_shared_global_8",
-      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.8",
-      "llvm_arguments": ["shared_ptr", "global_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.ca.shared.global.8 declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-all eight-byte call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 8."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"8"}]}
-    },
-    {
-      "id": "cp_async_ca_zfill_8",
-      "source_record": "int_nvvm_cp_async_ca_shared_global_8_s",
-      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.8.s",
-      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.ca.shared.global.8.s declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-all eight-byte source-size calls","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted both register and immediate source-size forms."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"8"},{"kind":"register_or_immediate"}]}
-    },
-    {
-      "id": "cp_async_ca_16",
-      "source_record": "int_nvvm_cp_async_ca_shared_global_16",
-      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.16",
-      "llvm_arguments": ["shared_ptr", "global_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.ca.shared.global.16 declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-all sixteen-byte call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 16."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"}]}
-    },
-    {
-      "id": "cp_async_ca_zfill_16",
-      "source_record": "int_nvvm_cp_async_ca_shared_global_16_s",
-      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.16.s",
-      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.ca.shared.global.16.s declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-all sixteen-byte source-size calls","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted both register and immediate source-size forms."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"},{"kind":"register_or_immediate"}]}
-    },
-    {
-      "id": "cp_async_cg_16",
-      "source_record": "int_nvvm_cp_async_cg_shared_global_16",
-      "llvm_symbol": "llvm.nvvm.cp.async.cg.shared.global.16",
-      "llvm_arguments": ["shared_ptr", "global_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.cg.shared.global.16 declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-global sixteen-byte call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.cg.shared.global with copy size 16."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","cg","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"}]}
-    },
-    {
-      "id": "cp_async_cg_zfill_16",
-      "source_record": "int_nvvm_cp_async_cg_shared_global_16_s",
-      "llvm_symbol": "llvm.nvvm.cp.async.cg.shared.global.16.s",
-      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.cg.shared.global.16.s declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes."},
-        {"targets":["sm_80","ptx70"],"representation":"typed AS3/AS1 cp.async cache-global sixteen-byte source-size calls","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted both register and immediate source-size forms."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","cg","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"},{"kind":"register_or_immediate"}]}
-    },
-    {
-      "id": "cp_async_commit_group",
-      "source_record": "int_nvvm_cp_async_commit_group",
-      "llvm_symbol": "llvm.nvvm.cp.async.commit.group",
-      "llvm_arguments": [],
-      "llvm_results": [],
-      "concrete_llvm_arguments": [],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.commit.group declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the zero-operand declaration and canonicalized it as nounwind."},
-        {"targets":["sm_80","ptx70"],"representation":"typed zero-operand cp.async commit-group call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.commit_group."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","commit_group"],"operands":[]}
-    },
-    {
-      "id": "cp_async_wait_all",
-      "source_record": "int_nvvm_cp_async_wait_all",
-      "llvm_symbol": "llvm.nvvm.cp.async.wait.all",
-      "llvm_arguments": [],
-      "llvm_results": [],
-      "concrete_llvm_arguments": [],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.wait.all declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the zero-operand declaration and canonicalized it as nounwind."},
-        {"targets":["sm_80","ptx70"],"representation":"typed zero-operand cp.async wait-all call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.wait_all."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","wait_all"],"operands":[]}
-    },
-    {
-      "id": "cp_async_wait_group",
-      "source_record": "int_nvvm_cp_async_wait_group",
-      "llvm_symbol": "llvm.nvvm.cp.async.wait.group",
-      "llvm_arguments": ["i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {"targets":["sm_80"],"representation":"typed llvm.nvvm.cp.async.wait.group(i32 immarg) declaration","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"LLVM recognized the declaration and canonicalized its operand as immarg and the function as nounwind."},
-        {"targets":["sm_80","ptx70"],"representation":"typed constant cp.async wait-group call","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"The selected Rust LLVM backend emitted exact cp.async.wait_group with an integer literal."},
-        {"targets":["sm_80","ptx70"],"representation":"PTX emitted by the selected Rust LLVM cp.async probe","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"CUDA 13.3 ptxas accepted the emitted PTX for sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic":"cp","modifiers":["async","wait_group"],"operands":[{"kind":"immediate"}]}
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-elect.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-elect.json
deleted file mode 100644
index b694d7cc93..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-elect.json
+++ /dev/null
@@ -1,77 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-elect",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {
-    "llvm_arguments": ["i32"],
-    "llvm_results": ["i32", "i1"],
-    "concrete_llvm_arguments": ["i32"],
-    "concrete_llvm_results": ["i32", "i1"],
-    "target_triple": "nvptx64-nvidia-cuda",
-    "gpu_target": "sm_90",
-    "ptx_feature": "+ptx80",
-    "status": "validated",
-    "declaration_attributes_canonicalized": true,
-    "runtime_validation": "unexecuted"
-  },
-  "fixtures": [
-    {
-      "id": "elect_sync_typed_probe",
-      "coverage_count": 1,
-      "stages": [
-        {
-          "targets": ["sm_90"],
-          "representation": "typed llvm.nvvm.elect.sync declaration and register/immediate calls",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as accepted the 1,080-byte source with SHA-256 e83b2cd7839b679e3d23da509c817d79108c84ce4eabefc70282dd2065cf69a1 and produced 2,100-byte bitcode with SHA-256 77043e86454445b633d31ec577a6c27c5a2675b960421a0739a621de3896a9e9."
-        },
-        {
-          "targets": ["sm_90", "ptx80"],
-          "representation": "typed LLVM probe retaining register and immediate member-mask forms",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM 22 emitted exactly elect.sync %r2|%p1, %r1 and elect.sync %r3|%p2, -1 with no extern calls; the 826-byte PTX had SHA-256 86e7fcebf74be1752a172688c87ec3bf042a296f7846df731b0385ad9145bf23."
-        },
-        {
-          "targets": ["sm_90", "ptx80"],
-          "representation": "LLVM-produced PTX retaining both elect.sync selections",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 4,328-byte cubin with SHA-256 f13f6d2c6aa0971a28ecf01f5008727bc513b0acef9cdd075973f03f30c15876.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    }
-  ],
-  "matrices": [
-    {
-      "axes": [{"name": "id", "values": ["elect_sync"]}],
-      "product_count": 1,
-      "fixtures": ["elect_sync_typed_probe"],
-      "template": {
-        "id": "${id}",
-        "source_record": "int_nvvm_elect_sync",
-        "llvm_symbol": "llvm.nvvm.elect.sync",
-        "expected_ptx": {
-          "mnemonic": "elect",
-          "modifiers": ["sync"],
-          "operands": [
-            {"kind": "register_predicate_pair"},
-            {"kind": "register_or_immediate"}
-          ]
-        },
-        "facts": {}
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-execution-control.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-execution-control.json
deleted file mode 100644
index f376b41d56..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-execution-control.json
+++ /dev/null
@@ -1,422 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-execution-control",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "barrier_cta_sync",
-      "source_record": "int_nvvm_barrier_cta_sync_count",
-      "llvm_symbol": "llvm.nvvm.barrier.cta.sync.count",
-      "llvm_arguments": ["i32", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_30",
-      "ptx_feature": "+ptx60",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_30"],
-          "representation": "typed counted-barrier declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 73bcf70a8647d141dea6d68aba4ab626818aef73c25eaf85b9c415814dcf0e71, bitcode SHA-256 50b842e412973300a1b0c67190c173f7f5f744db15efc9d639daa3b5c7ec33c6, and canonical IR SHA-256 d068e7a4ff32510dd439f5e92e97374a319e4ba6c1a6380cff5c855a0cde6a3f."
-        },
-        {
-          "targets": ["sm_30", "ptx60"],
-          "representation": "four register/immediate forms of barrier.sync",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 b220ca5fdae82f6048cc1e930fa07a9f6e8cf23a610476f0cf469a01e985a85d."
-        },
-        {
-          "targets": ["sm_75", "ptx60"],
-          "representation": "LLVM-produced PTX containing the four barrier.sync forms",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "barrier",
-        "modifiers": ["sync"],
-        "operands": [
-          {"kind": "register_or_immediate"},
-          {"kind": "register_or_immediate"}
-        ]
-      }
-    },
-    {
-      "id": "barrier_cta_sync_aligned",
-      "source_record": "int_nvvm_barrier_cta_sync_aligned_count",
-      "llvm_symbol": "llvm.nvvm.barrier.cta.sync.aligned.count",
-      "llvm_arguments": ["i32", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_20",
-      "ptx_feature": "+ptx32",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_20"],
-          "representation": "typed aligned counted-barrier declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 f5faf9f6ac1de5de821476c35d6cfed39ffb9abc1de2e7ac9038c6f51c7bda6c, bitcode SHA-256 3c90f0f9686014d1074c753f644ca9dac74397f465cb78f23e726820af47345d, and canonical IR SHA-256 70ee92254f4aa29db9731a3beddef77865f6f5bff7a81a7c66aa076726076ea4."
-        },
-        {
-          "targets": ["sm_20", "ptx32"],
-          "representation": "four register/immediate forms of bar.sync",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 d3fbf9d7aa025ee2f4836094b8652fb29a5557e94bbc73689c5b18dd04edc47a."
-        },
-        {
-          "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX containing the four bar.sync forms",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "bar",
-        "modifiers": ["sync"],
-        "operands": [
-          {"kind": "register_or_immediate"},
-          {"kind": "register_or_immediate"}
-        ]
-      }
-    },
-    {
-      "id": "barrier_cta_arrive",
-      "source_record": "int_nvvm_barrier_cta_arrive_count",
-      "llvm_symbol": "llvm.nvvm.barrier.cta.arrive.count",
-      "llvm_arguments": ["i32", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_30",
-      "ptx_feature": "+ptx60",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_30"],
-          "representation": "typed counted-barrier arrival declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 6844935a6b2fc799ecf3245780589f5de02583685caae0267be1f37668535a2b, bitcode SHA-256 d014888592c0d71d5c569711143349e9daaea3079578144385a41fa055e87992, and canonical IR SHA-256 e2a2a5029602f80fb55f6f90e96083d51ce76da3c019b666aad025d0e64c0987."
-        },
-        {
-          "targets": ["sm_30", "ptx60"],
-          "representation": "four register/immediate forms of barrier.arrive",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 94113e81da1085d1688c37f811dc497e0963be3af0f703f03862268e4a14678f."
-        },
-        {
-          "targets": ["sm_75", "ptx60"],
-          "representation": "LLVM-produced PTX containing the four barrier.arrive forms",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "barrier",
-        "modifiers": ["arrive"],
-        "operands": [
-          {"kind": "register_or_immediate"},
-          {"kind": "register_or_immediate"}
-        ]
-      }
-    },
-    {
-      "id": "barrier_cta_arrive_aligned",
-      "source_record": "int_nvvm_barrier_cta_arrive_aligned_count",
-      "llvm_symbol": "llvm.nvvm.barrier.cta.arrive.aligned.count",
-      "llvm_arguments": ["i32", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_20",
-      "ptx_feature": "+ptx32",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_20"],
-          "representation": "typed aligned counted-barrier arrival declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 042ab4e18a0df19c17f139123fd21759a9b179221f8d303f9d13c8079484f5cf, bitcode SHA-256 8c4a5ec77772763a7deb91899e5ae3a1fd10df067d619b985b4e6e971e8b2319, and canonical IR SHA-256 6ce3356d92a9740dfeb0f53cf4aa551586a1ee201a2210dadf631131e26ce047."
-        },
-        {
-          "targets": ["sm_20", "ptx32"],
-          "representation": "four register/immediate forms of bar.arrive",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted all four reviewed operand forms in PTX SHA-256 6b4d17807a12f8298fca5f4957b8e3da5ed27fcbd47111f8a173bbe61d5c8aad."
-        },
-        {
-          "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX containing the four bar.arrive forms",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "bar",
-        "modifiers": ["arrive"],
-        "operands": [
-          {"kind": "register_or_immediate"},
-          {"kind": "register_or_immediate"}
-        ]
-      }
-    },
-    {
-      "id": "grid_dependency_launch_dependents",
-      "source_record": "int_nvvm_griddepcontrol_launch_dependents",
-      "llvm_symbol": "llvm.nvvm.griddepcontrol.launch.dependents",
-      "llvm_arguments": [],
-      "llvm_results": [],
-      "concrete_llvm_arguments": [],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_90",
-      "ptx_feature": "+ptx78",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_90"],
-          "representation": "typed grid-dependency launch declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 73f4f353452e427acdb94ba15dbab4b8a6147137bf7e33decdbf8e41d03152e2, bitcode SHA-256 e666c4c454382bd351d83c3c8ca347bdd7d3bb9e24bb8e6643f306a4893f6ec1, and canonical IR SHA-256 7d1e8a9808ea508ca317b88b06284570bad89943a66e2dec4f7e8c7608688f94."
-        },
-        {
-          "targets": ["sm_90", "ptx78"],
-          "representation": "typed griddepcontrol.launch_dependents call",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted the exact instruction in PTX SHA-256 97504b6bc748ddc18d1e82bded39ecfefb288bf088e5afeca6c519f3a4f900cb."
-        },
-        {
-          "targets": ["sm_90", "ptx78"],
-          "representation": "LLVM-produced griddepcontrol.launch_dependents PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "griddepcontrol",
-        "modifiers": ["launch_dependents"],
-        "operands": []
-      }
-    },
-    {
-      "id": "grid_dependency_wait",
-      "source_record": "int_nvvm_griddepcontrol_wait",
-      "llvm_symbol": "llvm.nvvm.griddepcontrol.wait",
-      "llvm_arguments": [],
-      "llvm_results": [],
-      "concrete_llvm_arguments": [],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_90",
-      "ptx_feature": "+ptx78",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_90"],
-          "representation": "typed grid-dependency wait declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 d7f02e7d83c4693e1821d37b444a9dabb20710f703938b6bf6043b5f5bf48e90, bitcode SHA-256 f1cf4c3dca46407bea36ac7957f7da7909fa6405c8f6da0fc914ab6f5277ab9a, and canonical IR SHA-256 d2d4106872992bf9ca9d7b8b147a7a7ebc6a515b5ee748b9c1cb022eacd92450."
-        },
-        {
-          "targets": ["sm_90", "ptx78"],
-          "representation": "typed griddepcontrol.wait call",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted the exact instruction in PTX SHA-256 9b54ea0566c66befa6ff9e703c91eba251fa5c0fb9eb477a07c391fd9556045c."
-        },
-        {
-          "targets": ["sm_90", "ptx78"],
-          "representation": "LLVM-produced griddepcontrol.wait PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "griddepcontrol",
-        "modifiers": ["wait"],
-        "operands": []
-      }
-    },
-    {
-      "id": "setmaxnreg_inc",
-      "source_record": "int_nvvm_setmaxnreg_inc_sync_aligned_u32",
-      "llvm_symbol": "llvm.nvvm.setmaxnreg.inc.sync.aligned.u32",
-      "llvm_arguments": ["i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_90a",
-      "ptx_feature": "+ptx80",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_90a"],
-          "representation": "typed setmaxnreg increment declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 e451448a7385c879e0ca127c25cd038b92bffeacf94d6c3487dbeebbf4f2555e, bitcode SHA-256 00e92b75a1ef024a48fb31297bd9a46d831be64e916d3a522f16bc54d19046a3, and canonical IR SHA-256 7d3888169df2d273686c57ac7b169aaee07b0917f6db7b148a6403130aba3a68."
-        },
-        {
-          "targets": ["sm_90a", "ptx80"],
-          "representation": "typed setmaxnreg.inc.sync.aligned.u32 call",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted the exact instruction in PTX SHA-256 1212a8cf1cbba526030c0f67dbb854c5dd2d24e7cf7977ef2137a14638be0d8a."
-        },
-        {
-          "targets": ["sm_90a", "ptx80"],
-          "representation": "LLVM-produced setmaxnreg increment PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "setmaxnreg",
-        "modifiers": ["inc", "sync", "aligned", "u32"],
-        "operands": [{"kind": "immediate"}]
-      }
-    },
-    {
-      "id": "setmaxnreg_dec",
-      "source_record": "int_nvvm_setmaxnreg_dec_sync_aligned_u32",
-      "llvm_symbol": "llvm.nvvm.setmaxnreg.dec.sync.aligned.u32",
-      "llvm_arguments": ["i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_90a",
-      "ptx_feature": "+ptx80",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_90a"],
-          "representation": "typed setmaxnreg decrement declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized source IR SHA-256 a559d3e87b535201a9ae87b367b6bbc3cd6c6fc452574311161fa126b2cabfb7, bitcode SHA-256 6a87d7b67460deeb9fca1720371d2d9f1295620546fc167c64deee03d362543b, and canonical IR SHA-256 b5754b28a81283e193437f849a7e07af5488d484fcb5c3c3d2e9a2d15c193d2d."
-        },
-        {
-          "targets": ["sm_90a", "ptx80"],
-          "representation": "typed setmaxnreg.dec.sync.aligned.u32 call",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM emitted the exact instruction in PTX SHA-256 9758b60ad9ed87b7e0a90f9fd57c3edaa90c7c793621c5f911f0fcaef0a55555."
-        },
-        {
-          "targets": ["sm_90a", "ptx80"],
-          "representation": "LLVM-produced setmaxnreg decrement PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "setmaxnreg",
-        "modifiers": ["dec", "sync", "aligned", "u32"],
-        "operands": [{"kind": "immediate"}]
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-mbarrier-basic.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-mbarrier-basic.json
deleted file mode 100644
index 63f814ee45..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-mbarrier-basic.json
+++ /dev/null
@@ -1,258 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "mbarrier_init",
-      "source_record": "int_nvvm_mbarrier_init_shared",
-      "llvm_symbol": "llvm.nvvm.mbarrier.init.shared",
-      "llvm_arguments": ["shared_ptr", "i32"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "i32"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.mbarrier.init.shared declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed shared-memory mbarrier initialization",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact mbarrier.init.shared.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mbarrier",
-        "modifiers": ["init", "shared", "b64"],
-        "operands": [
-          { "kind": "address" },
-          { "kind": "register_or_immediate" }
-        ]
-      }
-    },
-    {
-      "id": "mbarrier_arrive",
-      "source_record": "int_nvvm_mbarrier_arrive_shared",
-      "llvm_symbol": "llvm.nvvm.mbarrier.arrive.shared",
-      "llvm_arguments": ["shared_ptr"],
-      "llvm_results": ["i64"],
-      "concrete_llvm_arguments": ["ptr addrspace(3)"],
-      "concrete_llvm_results": ["i64"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.mbarrier.arrive.shared declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed shared-memory mbarrier arrival",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact mbarrier.arrive.shared.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mbarrier",
-        "modifiers": ["arrive", "shared", "b64"],
-        "operands": [
-          { "kind": "register" },
-          { "kind": "address" }
-        ]
-      }
-    },
-    {
-      "id": "mbarrier_arrive_no_complete",
-      "source_record": "int_nvvm_mbarrier_arrive_noComplete_shared",
-      "llvm_symbol": "llvm.nvvm.mbarrier.arrive.noComplete.shared",
-      "llvm_arguments": ["shared_ptr", "i32"],
-      "llvm_results": ["i64"],
-      "concrete_llvm_arguments": ["ptr addrspace(3)", "i32"],
-      "concrete_llvm_results": ["i64"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.mbarrier.arrive.noComplete.shared declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the counted shared-pointer declaration with convergent and nocallback attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed counted shared-memory mbarrier no-complete arrival",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact mbarrier.arrive.noComplete.shared.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mbarrier",
-        "modifiers": ["arrive", "noComplete", "shared", "b64"],
-        "operands": [
-          { "kind": "register" },
-          { "kind": "address" },
-          { "kind": "register" }
-        ]
-      }
-    },
-    {
-      "id": "mbarrier_test_wait",
-      "source_record": "int_nvvm_mbarrier_test_wait_shared",
-      "llvm_symbol": "llvm.nvvm.mbarrier.test.wait.shared",
-      "llvm_arguments": ["shared_ptr", "i64"],
-      "llvm_results": ["i1"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "convergent side-effecting inline PTX mbarrier test-wait",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend retained exact mbarrier.test_wait.shared.b64 PTX and the predicate-to-i1 adapter."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mbarrier",
-        "modifiers": ["test_wait", "shared", "b64"],
-        "operands": [
-          { "kind": "register" },
-          { "kind": "address" },
-          { "kind": "register" }
-        ]
-      }
-    },
-    {
-      "id": "mbarrier_inval",
-      "source_record": "int_nvvm_mbarrier_inval_shared",
-      "llvm_symbol": "llvm.nvvm.mbarrier.inval.shared",
-      "llvm_arguments": ["shared_ptr"],
-      "llvm_results": [],
-      "concrete_llvm_arguments": ["ptr addrspace(3)"],
-      "concrete_llvm_results": [],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx70",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.mbarrier.inval.shared declaration",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized the declaration with convergent, nocallback, and argument-memory write attributes."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "typed shared-memory mbarrier invalidation",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact mbarrier.inval.shared.b64 PTX."
-        },
-        {
-          "targets": ["sm_80", "ptx70"],
-          "representation": "PTX emitted by the selected Rust LLVM mbarrier probe",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mbarrier",
-        "modifiers": ["inval", "shared", "b64"],
-        "operands": [{ "kind": "address" }]
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-packed-conversion-fp8-f16x2.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-packed-conversion-fp8-f16x2.json
deleted file mode 100644
index 8e0e86cbd3..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-packed-conversion-fp8-f16x2.json
+++ /dev/null
@@ -1,850 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "cvt_rn_satfinite_e4m3x2_f16x2",
-      "source_record": "int_nvvm_f16x2_to_e4m3x2_rn",
-      "llvm_symbol": "llvm.nvvm.f16x2.to.e4m3x2.rn",
-      "llvm_arguments": [
-        "v2f16"
-      ],
-      "llvm_results": [
-        "i16"
-      ],
-      "concrete_llvm_arguments": [
-        "v2f16"
-      ],
-      "concrete_llvm_results": [
-        "i16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "satfinite",
-          "e4m3x2",
-          "f16x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_satfinite_relu_e4m3x2_f16x2",
-      "source_record": "int_nvvm_f16x2_to_e4m3x2_rn_relu",
-      "llvm_symbol": "llvm.nvvm.f16x2.to.e4m3x2.rn.relu",
-      "llvm_arguments": [
-        "v2f16"
-      ],
-      "llvm_results": [
-        "i16"
-      ],
-      "concrete_llvm_arguments": [
-        "v2f16"
-      ],
-      "concrete_llvm_results": [
-        "i16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "satfinite",
-          "relu",
-          "e4m3x2",
-          "f16x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_satfinite_e5m2x2_f16x2",
-      "source_record": "int_nvvm_f16x2_to_e5m2x2_rn",
-      "llvm_symbol": "llvm.nvvm.f16x2.to.e5m2x2.rn",
-      "llvm_arguments": [
-        "v2f16"
-      ],
-      "llvm_results": [
-        "i16"
-      ],
-      "concrete_llvm_arguments": [
-        "v2f16"
-      ],
-      "concrete_llvm_results": [
-        "i16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "satfinite",
-          "e5m2x2",
-          "f16x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_satfinite_relu_e5m2x2_f16x2",
-      "source_record": "int_nvvm_f16x2_to_e5m2x2_rn_relu",
-      "llvm_symbol": "llvm.nvvm.f16x2.to.e5m2x2.rn.relu",
-      "llvm_arguments": [
-        "v2f16"
-      ],
-      "llvm_results": [
-        "i16"
-      ],
-      "concrete_llvm_arguments": [
-        "v2f16"
-      ],
-      "concrete_llvm_results": [
-        "i16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "satfinite",
-          "relu",
-          "e5m2x2",
-          "f16x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_f16x2_e4m3x2",
-      "source_record": "int_nvvm_e4m3x2_to_f16x2_rn",
-      "llvm_symbol": "llvm.nvvm.e4m3x2.to.f16x2.rn",
-      "llvm_arguments": [
-        "i16"
-      ],
-      "llvm_results": [
-        "v2f16"
-      ],
-      "concrete_llvm_arguments": [
-        "i16"
-      ],
-      "concrete_llvm_results": [
-        "v2f16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "f16x2",
-          "e4m3x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_relu_f16x2_e4m3x2",
-      "source_record": "int_nvvm_e4m3x2_to_f16x2_rn_relu",
-      "llvm_symbol": "llvm.nvvm.e4m3x2.to.f16x2.rn.relu",
-      "llvm_arguments": [
-        "i16"
-      ],
-      "llvm_results": [
-        "v2f16"
-      ],
-      "concrete_llvm_arguments": [
-        "i16"
-      ],
-      "concrete_llvm_results": [
-        "v2f16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "relu",
-          "f16x2",
-          "e4m3x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_f16x2_e5m2x2",
-      "source_record": "int_nvvm_e5m2x2_to_f16x2_rn",
-      "llvm_symbol": "llvm.nvvm.e5m2x2.to.f16x2.rn",
-      "llvm_arguments": [
-        "i16"
-      ],
-      "llvm_results": [
-        "v2f16"
-      ],
-      "concrete_llvm_arguments": [
-        "i16"
-      ],
-      "concrete_llvm_results": [
-        "v2f16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "f16x2",
-          "e5m2x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    },
-    {
-      "id": "cvt_rn_relu_f16x2_e5m2x2",
-      "source_record": "int_nvvm_e5m2x2_to_f16x2_rn_relu",
-      "llvm_symbol": "llvm.nvvm.e5m2x2.to.f16x2.rn.relu",
-      "llvm_arguments": [
-        "i16"
-      ],
-      "llvm_results": [
-        "v2f16"
-      ],
-      "concrete_llvm_arguments": [
-        "i16"
-      ],
-      "concrete_llvm_results": [
-        "v2f16"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_89",
-      "ptx_feature": "+ptx81",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "canonical LLVM declarations for all eight f16x2 and packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452)."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM intrinsic calls on a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "pure inline PTX with a single packed source operand",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking."
-        },
-        {
-          "targets": [
-            "sm_89",
-            "ptx81"
-          ],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "declaration_attributes_canonicalized": true,
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "cvt",
-        "modifiers": [
-          "rn",
-          "relu",
-          "f16x2",
-          "e5m2x2"
-        ],
-        "operands": [
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "register"
-          }
-        ]
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-packed-conversion-fp8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-packed-conversion-fp8.json
deleted file mode 100644
index eae42cd07e..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-packed-conversion-fp8.json
+++ /dev/null
@@ -1,120 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {
-    "llvm_arguments": ["f32", "f32"],
-    "llvm_results": ["i16"],
-    "concrete_llvm_arguments": ["f32", "f32"],
-    "concrete_llvm_results": ["i16"],
-    "target_triple": "nvptx64-nvidia-cuda",
-    "gpu_target": "sm_89",
-    "ptx_feature": "+ptx81",
-    "status": "validated",
-    "declaration_attributes_canonicalized": true,
-    "runtime_validation": "unexecuted"
-  },
-  "fixtures": [
-    {
-      "id": "fp8_conversion_sm89_ptx81",
-      "coverage_count": 4,
-      "stages": [
-        {
-          "targets": ["sm_89", "ptx81"],
-          "representation": "canonical LLVM declarations for all four packed-FP8 conversions",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a)."
-        },
-        {
-          "targets": ["sm_89", "ptx81"],
-          "representation": "typed LLVM intrinsic calls in PTX high, low source order",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected."
-        },
-        {
-          "targets": ["sm_89", "ptx81"],
-          "representation": "typed LLVM output assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": ["sm_89", "ptx81"],
-          "representation": "pure inline PTX with u16 output and public low, high inputs",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints."
-        },
-        {
-          "targets": ["sm_89", "ptx81"],
-          "representation": "inline PTX assembled at the exact floor",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    }
-  ],
-  "matrices": [
-    {
-      "axes": [
-        {"name": "format", "values": ["e4m3x2", "e5m2x2"]}
-      ],
-      "product_count": 2,
-      "fixtures": ["fp8_conversion_sm89_ptx81"],
-      "template": {
-        "id": "cvt_rn_satfinite_${format}_f32",
-        "source_record": "int_nvvm_ff_to_${format}_rn",
-        "llvm_symbol": "llvm.nvvm.ff.to.${format}.rn",
-        "expected_ptx": {
-          "mnemonic": "cvt",
-          "modifiers": ["rn", "satfinite", "${format}", "f32"],
-          "operands": [
-            {"kind": "register"},
-            {"kind": "register"},
-            {"kind": "register"}
-          ]
-        }
-      }
-    },
-    {
-      "axes": [
-        {"name": "format", "values": ["e4m3x2", "e5m2x2"]}
-      ],
-      "product_count": 2,
-      "fixtures": ["fp8_conversion_sm89_ptx81"],
-      "template": {
-        "id": "cvt_rn_satfinite_relu_${format}_f32",
-        "source_record": "int_nvvm_ff_to_${format}_rn_relu",
-        "llvm_symbol": "llvm.nvvm.ff.to.${format}.rn.relu",
-        "expected_ptx": {
-          "mnemonic": "cvt",
-          "modifiers": ["rn", "satfinite", "relu", "${format}", "f32"],
-          "operands": [
-            {"kind": "register"},
-            {"kind": "register"},
-            {"kind": "register"}
-          ]
-        }
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f16.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f16.json
deleted file mode 100644
index a81a8fbab7..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f16.json
+++ /dev/null
@@ -1,329 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {
-    "llvm_arguments": [
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "v2f16",
-      "v2f16"
-    ],
-    "llvm_results": [
-      "v2f16",
-      "v2f16"
-    ],
-    "target_triple": "nvptx64-nvidia-cuda",
-    "gpu_target": "sm_120a",
-    "ptx_feature": "+ptx87",
-    "status": "validated",
-    "runtime_validation": "unexecuted"
-  },
-  "fixtures": [
-    {
-      "id": "dense_f8f6f4_f16_all_targets",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "typed declaration comparison",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "typed convergence comparison",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "exact dense kind::f8f6f4 F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx88"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live."
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::f8f6f4 F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx88"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live."
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::f8f6f4 F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx88"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live."
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::f8f6f4 F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_100a",
-            "ptx86"
-          ],
-          "representation": "lower-PTX and disallowed-target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_110a",
-            "ptx90"
-          ],
-          "representation": "disallowed-target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_120",
-            "ptx87"
-          ],
-          "representation": "missing exact-target suffix boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121",
-            "ptx88"
-          ],
-          "representation": "missing exact-target suffix boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx87"
-          ],
-          "representation": "lower-PTX target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx87"
-          ],
-          "representation": "lower-PTX target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx87"
-          ],
-          "representation": "lower-PTX target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    }
-  ],
-  "matrices": [
-    {
-      "axes": [
-        {
-          "name": "a_element",
-          "values": [
-            "e2m1",
-            "e2m3",
-            "e3m2",
-            "e4m3",
-            "e5m2"
-          ]
-        },
-        {
-          "name": "b_element",
-          "values": [
-            "e2m1",
-            "e2m3",
-            "e3m2",
-            "e4m3",
-            "e5m2"
-          ]
-        }
-      ],
-      "product_count": 25,
-      "fixtures": [
-        "dense_f8f6f4_f16_all_targets"
-      ],
-      "template": {
-        "id": "mma_m16n8k32_f16_${a_element}_${b_element}",
-        "source_record": "int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_${a_element}_${b_element}_f16",
-        "llvm_symbol": "llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.${a_element}.${b_element}.f16",
-        "expected_ptx": {
-          "mnemonic": "mma",
-          "modifiers": [
-            "sync",
-            "aligned",
-            "m16n8k32",
-            "row",
-            "col",
-            "kind::f8f6f4",
-            "f16",
-            "${a_element}",
-            "${b_element}",
-            "f16"
-          ],
-          "operands": [
-            {
-              "kind": "register_list",
-              "length": 2
-            },
-            {
-              "kind": "register_list",
-              "length": 4
-            },
-            {
-              "kind": "register_list",
-              "length": 2
-            },
-            {
-              "kind": "register_list",
-              "length": 2
-            }
-          ]
-        }
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f32.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f32.json
deleted file mode 100644
index 3338ae031b..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-f8f6f4-f32.json
+++ /dev/null
@@ -1,333 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {
-    "llvm_arguments": [
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "f32",
-      "f32",
-      "f32",
-      "f32"
-    ],
-    "llvm_results": [
-      "f32",
-      "f32",
-      "f32",
-      "f32"
-    ],
-    "target_triple": "nvptx64-nvidia-cuda",
-    "gpu_target": "sm_120a",
-    "ptx_feature": "+ptx87",
-    "status": "validated",
-    "runtime_validation": "unexecuted"
-  },
-  "fixtures": [
-    {
-      "id": "dense_f8f6f4_f32_all_targets",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "typed declaration comparison",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "typed convergence comparison",
-          "stage": "backend_codegen",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "exact dense kind::f8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx88"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms."
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::f8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx88"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms."
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::f8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx88"
-          ],
-          "representation": "convergent dense kind::f8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms."
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::f8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_100a",
-            "ptx86"
-          ],
-          "representation": "lower-PTX and disallowed-target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_110a",
-            "ptx90"
-          ],
-          "representation": "disallowed-target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_120",
-            "ptx87"
-          ],
-          "representation": "missing exact-target suffix boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121",
-            "ptx88"
-          ],
-          "representation": "missing exact-target suffix boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx87"
-          ],
-          "representation": "lower-PTX target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx87"
-          ],
-          "representation": "lower-PTX target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx87"
-          ],
-          "representation": "lower-PTX target boundary",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    }
-  ],
-  "matrices": [
-    {
-      "axes": [
-        {
-          "name": "a_element",
-          "values": [
-            "e2m1",
-            "e2m3",
-            "e3m2",
-            "e4m3",
-            "e5m2"
-          ]
-        },
-        {
-          "name": "b_element",
-          "values": [
-            "e2m1",
-            "e2m3",
-            "e3m2",
-            "e4m3",
-            "e5m2"
-          ]
-        }
-      ],
-      "product_count": 25,
-      "fixtures": [
-        "dense_f8f6f4_f32_all_targets"
-      ],
-      "template": {
-        "id": "mma_m16n8k32_f32_${a_element}_${b_element}",
-        "source_record": "int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_${a_element}_${b_element}_f32",
-        "llvm_symbol": "llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.${a_element}.${b_element}.f32",
-        "expected_ptx": {
-          "mnemonic": "mma",
-          "modifiers": [
-            "sync",
-            "aligned",
-            "m16n8k32",
-            "row",
-            "col",
-            "kind::f8f6f4",
-            "f32",
-            "${a_element}",
-            "${b_element}",
-            "f32"
-          ],
-          "operands": [
-            {
-              "kind": "register_list",
-              "length": 4
-            },
-            {
-              "kind": "register_list",
-              "length": 4
-            },
-            {
-              "kind": "register_list",
-              "length": 2
-            },
-            {
-              "kind": "register_list",
-              "length": 4
-            }
-          ]
-        }
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-mxf8f6f4-f32.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-mxf8f6f4-f32.json
deleted file mode 100644
index 3f6bced419..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-mxf8f6f4-f32.json
+++ /dev/null
@@ -1,235 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {
-    "llvm_arguments": [
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "i32",
-      "f32",
-      "f32",
-      "f32",
-      "f32",
-      "i32",
-      "i16",
-      "i16",
-      "i32",
-      "i16",
-      "i16"
-    ],
-    "llvm_results": [
-      "f32",
-      "f32",
-      "f32",
-      "f32"
-    ],
-    "target_triple": "nvptx64-nvidia-cuda",
-    "gpu_target": "sm_120a",
-    "ptx_feature": "+ptx87",
-    "status": "validated",
-    "runtime_validation": "unexecuted"
-  },
-  "fixtures": [
-    {
-      "id": "dense_mxf8f6f4_f32_all_targets",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16."
-        },
-        {
-          "targets": [
-            "sm_120a",
-            "ptx87"
-          ],
-          "representation": "exact dense kind::mxf8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx88"
-          ],
-          "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d."
-        },
-        {
-          "targets": [
-            "sm_120f",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::mxf8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx88"
-          ],
-          "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8."
-        },
-        {
-          "targets": [
-            "sm_121a",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::mxf8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx88"
-          ],
-          "representation": "generated dense kind::mxf8f6f4 F32 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c."
-        },
-        {
-          "targets": [
-            "sm_121f",
-            "ptx88"
-          ],
-          "representation": "exact dense kind::mxf8f6f4 F32 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
-        }
-      ]
-    }
-  ],
-  "matrices": [
-    {
-      "axes": [
-        {
-          "name": "a_element",
-          "values": [
-            "e2m1",
-            "e2m3",
-            "e3m2",
-            "e4m3",
-            "e5m2"
-          ]
-        },
-        {
-          "name": "b_element",
-          "values": [
-            "e2m1",
-            "e2m3",
-            "e3m2",
-            "e4m3",
-            "e5m2"
-          ]
-        }
-      ],
-      "product_count": 25,
-      "fixtures": [
-        "dense_mxf8f6f4_f32_all_targets"
-      ],
-      "template": {
-        "id": "mma_m16n8k32_mxf8f6f4_f32_${a_element}_${b_element}",
-        "source_record": "int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_${a_element}_${b_element}_f32_ue8m0",
-        "llvm_symbol": "llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.${a_element}.${b_element}.f32.ue8m0",
-        "expected_ptx": {
-          "mnemonic": "mma",
-          "modifiers": [
-            "sync",
-            "aligned",
-            "m16n8k32",
-            "row",
-            "col",
-            "kind::mxf8f6f4",
-            "block_scale",
-            "f32",
-            "${a_element}",
-            "${b_element}",
-            "f32",
-            "ue8m0"
-          ],
-          "operands": [
-            {
-              "kind": "register_list",
-              "length": 4
-            },
-            {
-              "kind": "register_list",
-              "length": 4
-            },
-            {
-              "kind": "register_list",
-              "length": 2
-            },
-            {
-              "kind": "register_list",
-              "length": 4
-            },
-            {
-              "kind": "register"
-            },
-            {
-              "kind": "register_list",
-              "length": 2
-            },
-            {
-              "kind": "register"
-            },
-            {
-              "kind": "register_list",
-              "length": 2
-            }
-          ]
-        }
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-int8-m16n8k64.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-int8-m16n8k64.json
deleted file mode 100644
index b28436afda..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-int8-m16n8k64.json
+++ /dev/null
@@ -1,146 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "mma_sp_m16n8k64_s32_s8",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 signed INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping s8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_s8_u8",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s8_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s8.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping s8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u8",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 unsigned INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping u8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u8_s8",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u8_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u8.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping u8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_s8_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 signed saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating s8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_s8_u8_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating s8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u8_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 unsigned saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating u8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u8_s8_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx71"], "representation": "convergent base sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating u8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx71"], "representation": "exact-floor base sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f16.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f16.json
deleted file mode 100644
index abb1758e73..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f16.json
+++ /dev/null
@@ -1,163 +0,0 @@
-{
-  "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {
-    "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "v2f16", "v2f16", "i32", "i32"],
-    "llvm_results": ["v2f16", "v2f16"],
-    "target_triple": "nvptx64-nvidia-cuda",
-    "gpu_target": "sm_120a",
-    "ptx_feature": "+ptx87",
-    "status": "validated",
-    "runtime_validation": "unexecuted"
-  },
-  "fixtures": [
-    {
-      "id": "ordered_f8f6f4_f16_sm120a_ptx87",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": ["sm_120a", "ptx87"],
-          "representation": "typed declaration and convergence comparison",
-          "stage": "declaration_canonicalization",
-          "mechanism": "typed_nvvm",
-          "outcome": "succeeded",
-          "detail": "The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier."
-        },
-        {
-          "targets": ["sm_120a", "ptx87"],
-          "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX in public C, A, B, metadata, selector order",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls."
-        },
-        {
-          "targets": ["sm_120a", "ptx87"],
-          "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    },
-    {
-      "id": "ordered_f8f6f4_f16_sm120f_ptx88",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": ["sm_120f", "ptx88"],
-          "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls."
-        },
-        {
-          "targets": ["sm_120f", "ptx88"],
-          "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    },
-    {
-      "id": "ordered_f8f6f4_f16_sm121a_ptx88",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": ["sm_121a", "ptx88"],
-          "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls."
-        },
-        {
-          "targets": ["sm_121a", "ptx88"],
-          "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    },
-    {
-      "id": "ordered_f8f6f4_f16_sm121f_ptx88",
-      "coverage_count": 25,
-      "stages": [
-        {
-          "targets": ["sm_121f", "ptx88"],
-          "representation": "convergent ordered kind::f8f6f4 packed-F16 inline PTX",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls."
-        },
-        {
-          "targets": ["sm_121f", "ptx88"],
-          "representation": "exact-floor ordered kind::f8f6f4 packed-F16 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms.",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ]
-    }
-  ],
-  "matrices": [
-    {
-      "axes": [
-        {"name": "a_element", "values": ["e2m1", "e2m3", "e3m2", "e4m3", "e5m2"]},
-        {"name": "b_element", "values": ["e2m1", "e2m3", "e3m2", "e4m3", "e5m2"]}
-      ],
-      "product_count": 25,
-      "fixtures": [
-        "ordered_f8f6f4_f16_sm120a_ptx87",
-        "ordered_f8f6f4_f16_sm120f_ptx88",
-        "ordered_f8f6f4_f16_sm121a_ptx88",
-        "ordered_f8f6f4_f16_sm121f_ptx88"
-      ],
-      "template": {
-        "id": "mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_${a_element}_${b_element}_f16",
-        "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_${a_element}_${b_element}_f16",
-        "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.${a_element}.${b_element}.f16",
-        "expected_ptx": {
-          "mnemonic": "mma",
-          "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "kind::f8f6f4", "f16", "${a_element}", "${b_element}", "f16"],
-          "operands": [
-            {"kind": "register_list", "length": 2},
-            {"kind": "register_list", "length": 4},
-            {"kind": "register_list", "length": 4},
-            {"kind": "register_list", "length": 2},
-            {"kind": "register"},
-            {"kind": "immediate"}
-          ]
-        }
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k128.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k128.json
deleted file mode 100644
index dc5d4ea417..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k128.json
+++ /dev/null
@@ -1,146 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 5,008-byte combined inline-PTX probe (SHA-256 7145b7febe8ce9f319a106e508a95639d3c06ac7a7cdf1be2c17e0c1f637e516) used public C, A, B, metadata, selector order and covered all eight variants with selector 0. Pinned LLVM emitted 4,025-byte PTX (SHA-256 9eda0f598c87254794f6e3d9e4cae4477b86b5ccb2a0146909c1fed715cce198) with exactly eight ordered m16n8k128 INT4 instructions. An auxiliary 4,800-byte typed LLVM comparison (SHA-256 71c233f005c86cc6616c05c5189171fc62d1525e6e8509927c194177935e8778) used source A, B, C, metadata, selector order and emitted 13,746-byte PTX (SHA-256 5d62e76cc6106809b4a017d82b2b8191dff13107effc9908de13b3a18406eb30) with the same eight forms; this record retained wrapping s4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted all eight instructions and produced a 4,008-byte cubin with SHA-256 bdab25ef1571d327ffd4bb0023276418b0f44548683778242827997a726ac660 containing exactly eight IMMA.SP.168128 instructions. LLVM and ptxas rejected selector 1 and dynamic selectors; ptxas also rejected PTX 8.4 and sm_75.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4_u4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping s4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping u4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4_s4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping u4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating s4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating s4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating u4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k128 INT4 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating u4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k128 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k64.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k64.json
deleted file mode 100644
index 8b05bbb497..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int4-m16n8k64.json
+++ /dev/null
@@ -1,146 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 8,107-byte combined inline-PTX probe (SHA-256 ce31bb4c5d7f083ae8e1293a0a3e5196575482427a041d549f9ecd3862e46942) used public C, A, B, metadata, selector order and covered all eight variants with both selector values. Pinned LLVM emitted 8,453-byte PTX (SHA-256 7a8bb0261fd16c1941b47500d70f3cff95a1acb95c51e52480af15a4effc8818) with exactly 16 ordered m16n8k64 INT4 instructions. An auxiliary 6,948-byte typed LLVM comparison (SHA-256 379488c94c4e3d6b1806928d72723845b7bd329857f4dfc3caf8f74d7a5c6649) used source A, B, C, metadata, selector order and emitted 22,330-byte PTX (SHA-256 913b74cc0dba68779655073b83e5a534bcf076fa5e6a1d1d6029ef951b0fc1cd) with the same 16 forms; this record retained wrapping s4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted all 16 instructions and produced a 4,008-byte cubin with SHA-256 a6df23e0492e6f26ede3da7f7fab6c9fdab2693c5c2d6bda12339948cba640ec containing exactly 16 IMMA.SP.16864 instructions. LLVM and ptxas rejected selector 2 and dynamic selectors; ptxas also rejected PTX 8.4 and sm_75.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4_u4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping s4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping u4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4_s4",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping u4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating s4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating s4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating u4-by-u4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4.s4",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 INT4 inline PTX with selectors 0 and 1", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating u4-by-s4 PTX."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT4 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8-m16n8k64.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8-m16n8k64.json
deleted file mode 100644
index 2b9e3448d8..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8-m16n8k64.json
+++ /dev/null
@@ -1,146 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 signed INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping s8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8_u8",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping s8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 unsigned INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping u8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8_s8",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 mixed-sign INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping u8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 signed saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating s8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating s8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 unsigned saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating u8-by-u8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    },
-    {
-      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite",
-      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8_s8",
-      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8.s8",
-      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
-      "llvm_results": ["i32", "i32", "i32", "i32"],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx85",
-      "status": "validated",
-      "stages": [
-        {"targets": ["sm_80", "ptx85"], "representation": "convergent ordered sparse m16n8k64 mixed-sign saturating INT8 inline PTX with selector 0", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating u8-by-s8 PTX with selector 0."},
-        {"targets": ["sm_80", "ptx85"], "representation": "exact-floor ordered sparse m16n8k64 INT8 PTX", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"}
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-standard-int4.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-standard-int4.json
deleted file mode 100644
index 572cd5b230..0000000000
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-standard-int4.json
+++ /dev/null
@@ -1,1570 +0,0 @@
-{
-  "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
-  "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "records": [
-    {
-      "id": "mma_sp_m16n8k64_s32_s4",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 wrapping s4-by-s4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "s32",
-          "s4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_s4_u4",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s4.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 wrapping s4-by-u4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "s32",
-          "s4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u4",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 wrapping u4-by-u4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "s32",
-          "u4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u4_s4",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u4.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 wrapping u4-by-s4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "s32",
-          "u4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 saturating s4-by-s4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "s4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_s4_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 saturating s4-by-u4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "s4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 saturating u4-by-u4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "u4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k64_s32_u4_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k64 saturating u4-by-s4 INT4 inline PTX with selectors 0 and 1",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k64 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k64",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "u4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 2
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_s4",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 wrapping s4-by-s4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "s32",
-          "s4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_s4_u4",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.s4.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 wrapping s4-by-u4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "s32",
-          "s4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_u4",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 wrapping u4-by-u4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "s32",
-          "u4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_u4_s4",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.u4.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 wrapping u4-by-s4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "s32",
-          "u4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 saturating s4-by-s4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "s4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_s4_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 saturating s4-by-u4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "s4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_u4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 saturating u4-by-u4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "u4",
-          "u4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    },
-    {
-      "id": "mma_sp_m16n8k128_s32_u4_s4_satfinite",
-      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4_s4",
-      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4.s4",
-      "llvm_arguments": [
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "llvm_results": [
-        "i32",
-        "i32",
-        "i32",
-        "i32"
-      ],
-      "target_triple": "nvptx64-nvidia-cuda",
-      "gpu_target": "sm_80",
-      "ptx_feature": "+ptx71",
-      "status": "validated",
-      "stages": [
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "convergent standard sparse m16n8k128 saturating u4-by-s4 INT4 inline PTX with selector 0",
-          "stage": "backend_codegen",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "detail": "The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX."
-        },
-        {
-          "targets": [
-            "sm_80",
-            "ptx71"
-          ],
-          "representation": "exact-floor standard sparse m16n8k128 INT4 PTX",
-          "stage": "ptx_assembly",
-          "mechanism": "inline_ptx",
-          "outcome": "succeeded",
-          "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9).",
-          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
-        }
-      ],
-      "runtime_validation": "unexecuted",
-      "expected_ptx": {
-        "mnemonic": "mma",
-        "modifiers": [
-          "sp",
-          "sync",
-          "aligned",
-          "m16n8k128",
-          "row",
-          "col",
-          "satfinite",
-          "s32",
-          "u4",
-          "s4",
-          "s32"
-        ],
-        "operands": [
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register_list",
-            "length": 4
-          },
-          {
-            "kind": "register"
-          },
-          {
-            "kind": "immediate"
-          }
-        ]
-      }
-    }
-  ]
-}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-blackwell-ldmatrix.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix.json
similarity index 55%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-blackwell-ldmatrix.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix.json
index c3e2fe5c5f..a405e9a346 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-blackwell-ldmatrix.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [
       "anyptr"
@@ -29,271 +29,271 @@
             "sm_100a",
             "ptx86"
           ],
-          "representation": "one opaque-pointer LLVM module containing all 12 exact typed Blackwell ldmatrix calls",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae."
         },
         {
           "targets": [
             "sm_100a",
             "ptx86"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100a",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028."
         },
         {
           "targets": [
             "sm_100f",
             "ptx88"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_100f",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088."
         },
         {
           "targets": [
             "sm_103a",
             "ptx88"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103a",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7."
         },
         {
           "targets": [
             "sm_103f",
             "ptx88"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_103f",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f."
         },
         {
           "targets": [
             "sm_110a",
             "ptx90"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110a",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8."
         },
         {
           "targets": [
             "sm_110f",
             "ptx90"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_110f",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b."
         },
         {
           "targets": [
             "sm_120a",
             "ptx87"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120a",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532."
         },
         {
           "targets": [
             "sm_120f",
             "ptx88"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_120f",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760."
         },
         {
           "targets": [
             "sm_121a",
             "ptx88"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121a",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63."
         },
         {
           "targets": [
             "sm_121f",
             "ptx88"
           ],
-          "representation": "LLVM-produced PTX containing all 12 exact Blackwell ldmatrix heads for sm_121f",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee."
         },
         {
           "targets": [
             "sm_100a",
             "ptx86"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100a",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_100f",
             "ptx88"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_100f",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_103a",
             "ptx88"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103a",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_103f",
             "ptx88"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_103f",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_110a",
             "ptx90"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110a",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_110f",
             "ptx90"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_110f",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_120a",
             "ptx87"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120a",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_120f",
             "ptx88"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_120f",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_121a",
             "ptx88"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121a",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_121f",
             "ptx88"
           ],
-          "representation": "CUDA-assembled LLVM PTX containing the combined 12-call fixture for sm_121f",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-clc.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-clc.json
similarity index 64%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-clc.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-clc.json
index 13b645a678..22ad02432a 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-clc.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-clc.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-clc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-clc",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_100",
@@ -20,31 +20,31 @@
       "stages": [
         {
           "targets": ["sm_100"],
-          "representation": "typed declarations for unicast try-cancel and all four query operations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b."
+          "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e."
         },
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "one typed LLVM kernel retaining unicast try-cancel and all four CLC queries",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996."
+          "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316."
         },
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "typed LLVM PTX retaining the unicast and query kernel",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d.",
+          "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -54,31 +54,31 @@
       "stages": [
         {
           "targets": ["sm_100a"],
-          "representation": "typed declaration for multicast try-cancel",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as accepted the 390-byte source with SHA-256 e4ed73cd8c39af86e8723ca03126c44f31cc9511a73b6ece710b7875bc128b14 and produced 1,912-byte bitcode with SHA-256 2d37df5745e6221f4b75791540de98edab835e3370f36043bcf6aa7af4a3c22d."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd29e45eadad739b58e43893296b4b8ac5e05d43c3c3298836fb126503d61e83."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "one typed LLVM kernel retaining multicast try-cancel",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM 22 emitted one exact multicast instruction with no extern call for sm_100a/PTX 8.6, sm_101a/PTX 8.6, sm_103a/PTX 8.8, sm_110a/PTX 9.0, sm_120a/PTX 8.7, and sm_121a/PTX 9.0. Their PTX SHA-256 values were 366b5155464356e2e01d29fb3ad209f365ba68fc2e540489ce7e2f5830658cde, bf925481f00d8023b5493cf2daee655ba90e32486d16b140c013df87bc1b5625, 6310890cf30d93ae0d02d63362350bfd13428ec4522358d80b459a3b1e86435b, 7ae967df1894e5a0158eabed03a607eeb971d1d2f7e5ad2892ca7f632d95ccd7, a4d5dd50691a5e0d3a32d4c77bf87258009ecebf2d66d00bb73292d0cbb4fa4e, and c52b501ba903c302697c994c9ea3abd4506b8a3aee1fd6ca320cad533140c894. Plain sm_100 failed instruction selection, as required."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 423445d4b9994ee20be92cbabc0ad7c8ea667c6360848df3b2a5be7020df4051."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "typed LLVM PTX retaining the multicast kernel",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas assembled the exact probes for sm_100a, sm_103a, sm_110a, sm_120a, and sm_121a. Their cubin SHA-256 values were 30160dcbe4de4c64e9b78f907e2663cdfb244fbe1a5dfbba66a6b8a1c52d5d76, 357d54085a3c661f8b0841d933d2d1600a688b54a5004ac0610469c35d016c57, 6d3afe7f862c5f2b5931972aa5c5dac9e2649f14407dfe67ade547b72af951db, e481feac9e4382025b1204c38614a1f5bdf141f9ce226da45d4ff835d58fdaba, and dcb515f3f0bb87fb6ffb2e74b78542979a0264f56b6df43a3913786e201c44f9. CUDA 13.3 does not expose the LLVM-only sm_101a name.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f8543fdf61a5a323f880af2fdfe1b86d2ee1311a024f5a9e88d6c40a23845925.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-barrier.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-barrier.json
similarity index 53%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-barrier.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-barrier.json
index f5374ae787..bfe11cabcf 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-barrier.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-barrier.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [],
     "llvm_results": [],
@@ -24,31 +24,31 @@
       "stages": [
         {
           "targets": ["sm_90"],
-          "representation": "typed declarations for the four PTX 7.8 cluster-barrier records",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "one typed probe retaining the four PTX 7.8 cluster-barrier instructions",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "LLVM-produced PTX containing the four PTX 7.8 cluster-barrier instructions",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -58,31 +58,31 @@
       "stages": [
         {
           "targets": ["sm_90"],
-          "representation": "typed declarations for the two PTX 8.0 relaxed cluster-barrier records",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as accepted the convergent nocallback declarations; source IR SHA-256 2e2c065b1abe60a62a61678dca92352084f523640ebf0feb0dd47e0f8417b18f and bitcode SHA-256 7522f3d31dd29407d8c5d8b4c35ad5f38f663067ad3c415effb5d096b4d93463."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c5b7ef9743a0c36a3ad27cf3093c5d3160111a0acee880922a93dab4ce330a7."
         },
         {
           "targets": ["sm_90", "ptx80"],
-          "representation": "one typed probe retaining the two PTX 8.0 relaxed cluster-barrier instructions",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM emitted each exact instruction once; PTX SHA-256 4e8650939b44a80b88fe91ede1a9568ac5526889254c9d6954d52176c10a9714."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 462bdc8a89623f986fa9c307a069d444805dcb8644ef7f9793f45a48a5cc94c6."
         },
         {
           "targets": ["sm_90", "ptx80"],
-          "representation": "LLVM-produced PTX containing the two PTX 8.0 relaxed cluster-barrier instructions",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced cubin SHA-256 698cf65618fd422030be4645284d6ab56b24f75d0e955a49380ef78bb8e044a8.",
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5501565e9a498f7f5dba887004bc2cf506c57249b8a4a2e8fdebcefc8007b146.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-memory.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-memory.json
similarity index 58%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-memory.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-memory.json
index 69f9e23897..178a308975 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-memory.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-memory.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-cluster-memory",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-cluster-memory",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_90",
@@ -19,23 +19,23 @@
       "stages": [
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "one convergent LLVM kernel containing exact mapa and remote u32-load inline PTX",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 690-byte source had SHA-256 ad8a94dfa560d523bfe8cf963f8779132c960e1e6e5717e080e72819936f446a; the 1,007-byte PTX had SHA-256 005a7f90c8a3bd97a99792ce8a181d43c28fb318cd20dce6068fcb349800a951."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418a46a6a0279c0e2c7391fcccae3b3bec1a660a3ffad28b1df2a7f63df297d6."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "LLVM-produced PTX retaining both cluster-memory operations",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,712-byte cubin with SHA-256 1780aee43cd691bd7e3a79f4bc4b2d163e5cdfc11e08441ae318dfb091a5057a.",
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bf4f5d63cf41eae708cc3087c76cbea8b9812ba26f1c915646ef4b0205c51760.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-sreg.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-sreg.json
similarity index 68%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-sreg.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-sreg.json
index e1a8e2e798..26e589e8f9 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-cluster-sreg.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cluster-sreg.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [],
     "llvm_results": ["i32"],
@@ -24,34 +24,34 @@
       "stages": [
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "typed LLVM cluster special-register declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized all declarations with noundef, range, and memory(none) attributes; canonical IR SHA-256 cc683476582e11f0c6a024e74f7e0f87ff167c992170fba3731d93aa8dfb331e."
+          "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 070eb640b89420af5d8d67631d8269d7b3efa14da6bad70274af322cf7c68ee4."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "typed LLVM calls lowered to exact mov.u32 special-register reads",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc lowered the 18-record audit fixture, including all 14 admitted records, to exact PTX (SHA-256 f432e117b592bf39cf1f570a64dae4cc1441ecb28067b58a5dc4cd6beec5bef6). sm_80 rejected selection and PTX 7.7 rejected the sm_90 target floor.",
-          "tool_path": "llc",
-          "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+          "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0059b9999da5a35149dd382b0eca010573789cfffa14abf3009e8bd4aa054eec.",
+          "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+          "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "exact-floor cluster special-register PTX",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin (SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843).",
+          "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06c87e1559d3544f2076dc749c8ca032775629f52dd5fe2995f0a93307534852.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async-mbarrier.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async-mbarrier.json
new file mode 100644
index 0000000000..0617e54b7e
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async-mbarrier.json
@@ -0,0 +1,206 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "cp_async_mbarrier_arrive",
+      "source_record": "int_nvvm_cp_async_mbarrier_arrive",
+      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive",
+      "llvm_arguments": ["ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3d6f3f642d9e80245cb2f6c76442ccd0d824326e288898d121d5645368edcc8e."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8b4dfc81ee704ec2a8f8e8502b5f5673cafe35607eb28120b57fff0c3b27baf0."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 86913ed12af6b0bd59e448e3b5ccc6823e2f0b6a140146ead578815ea6f7c5ec.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cp",
+        "modifiers": ["async", "mbarrier", "arrive", "b64"],
+        "operands": [{ "kind": "address" }]
+      }
+    },
+    {
+      "id": "cp_async_mbarrier_arrive_shared",
+      "source_record": "int_nvvm_cp_async_mbarrier_arrive_shared",
+      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive.shared",
+      "llvm_arguments": ["shared_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cd2056552d6e2b796c195f31ad768ca175eac447825d819f5957b9fcec07059c."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d59be5c8527b799671b85f003a16a51bbd07448cc2130863d25e7318f2841384."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 23bdc05d4be1c27246a496b4f2ebede69efa8b84d326395adc4d804f16f9eef5.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cp",
+        "modifiers": ["async", "mbarrier", "arrive", "shared", "b64"],
+        "operands": [{ "kind": "address" }]
+      }
+    },
+    {
+      "id": "cp_async_mbarrier_arrive_noinc",
+      "source_record": "int_nvvm_cp_async_mbarrier_arrive_noinc",
+      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive.noinc",
+      "llvm_arguments": ["ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11cd942627957bdbe07521415b591132bac6543a6d37a10286821ca4be498129."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d65ebc7c63cdfa57f0f5aefa206fd07f6004879beb44350640071d5eb97e6625."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ef8193aee88f5efa25293d77c1399be0b83dd294401561177e91f144714072a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cp",
+        "modifiers": ["async", "mbarrier", "arrive", "noinc", "b64"],
+        "operands": [{ "kind": "address" }]
+      }
+    },
+    {
+      "id": "cp_async_mbarrier_arrive_noinc_shared",
+      "source_record": "int_nvvm_cp_async_mbarrier_arrive_noinc_shared",
+      "llvm_symbol": "llvm.nvvm.cp.async.mbarrier.arrive.noinc.shared",
+      "llvm_arguments": ["shared_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b62ad0b051d6ccaa435c3686d1a94547c3635490996a003e78abb772b160cc4."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9affe7627501776e21eb8059bdb4bee726a956435720cc6de05ee4d93ef51be1."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 366217f18f9473cf5ccc1d87632a590d8fcf7a8f5508a5bf876c49df1dc939e2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cp",
+        "modifiers": ["async", "mbarrier", "arrive", "noinc", "shared", "b64"],
+        "operands": [{ "kind": "address" }]
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async.json
new file mode 100644
index 0000000000..9f5abdda55
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-cp-async.json
@@ -0,0 +1,241 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "cp_async_ca_4",
+      "source_record": "int_nvvm_cp_async_ca_shared_global_4",
+      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.4",
+      "llvm_arguments": ["shared_ptr", "global_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 62367bcc29688f386037306dee01e279908da96b4fbc872a0731f6c1efe3b979."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fc90fb194ad06cc37e4ec41495d29b2ced1dadd23700f4a6742ba5cb274f236c."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61893d39abe94a66af4946061499f48731945913886e7c08b8e5c3761e2e2089.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"4"}]}
+    },
+    {
+      "id": "cp_async_ca_zfill_4",
+      "source_record": "int_nvvm_cp_async_ca_shared_global_4_s",
+      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.4.s",
+      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7c8298238217b06520a14baa7dd9017ef6824f9a237ff1c1059c16bcf4a79797."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3022aaec048152a29d81d9b43ad7121669183c87def7ffd6bcc83d52d7040f91."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9928045f3f05dc39ca94c224c2afe7c21092310cdfa83a18da46628fa549c1cc.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"4"},{"kind":"register_or_immediate"}]}
+    },
+    {
+      "id": "cp_async_ca_8",
+      "source_record": "int_nvvm_cp_async_ca_shared_global_8",
+      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.8",
+      "llvm_arguments": ["shared_ptr", "global_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b0555ffe6d212958fd651503471d80d0a62030dc35f1dbdac5aedd702be82dd9."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4115fad6c387b8a02ba11bff217ad8d8f244b96f080aad856ea45b4475ee27f8."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2b95d90d47b38d58ceb6c07c46652eac5c8dec1e1ae2ede4f13972c17c1b268.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"8"}]}
+    },
+    {
+      "id": "cp_async_ca_zfill_8",
+      "source_record": "int_nvvm_cp_async_ca_shared_global_8_s",
+      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.8.s",
+      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0b171b000534f40633027c1b9b7ab26b739a7ebb686ec8617019f3834cad070."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a01e92a24c61a5a33a96cee208154315907b4538b97345be0e979839c5358c02."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa2dd28dc4c740dac8fce5c55f0002a8bf7e9ab2210bacfd152777eb0e1928b0.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"8"},{"kind":"register_or_immediate"}]}
+    },
+    {
+      "id": "cp_async_ca_16",
+      "source_record": "int_nvvm_cp_async_ca_shared_global_16",
+      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.16",
+      "llvm_arguments": ["shared_ptr", "global_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b7c596a91052b75958df78b963338ca733b9b53643eb4b669015b106f926a4."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 99c87c9201c30820920f63624477dfab9046044be0c1e4e9e878a8bd5edd425f."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d4109cb4c2deaedb98fce7f124afb56882fc7ca5ef239a53d86a293d152b465.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"}]}
+    },
+    {
+      "id": "cp_async_ca_zfill_16",
+      "source_record": "int_nvvm_cp_async_ca_shared_global_16_s",
+      "llvm_symbol": "llvm.nvvm.cp.async.ca.shared.global.16.s",
+      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4eb30d0f094641ac8dc5af009c2818cf66555f890dc7e6ed4c8e92987f6ee493."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91e8999a4e6f9db5fd5ba08aa79bd58870a77e05e12b06523f182ec6cfb7a14d."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d5bc6dde1a35b6d38bf05e0d3aea8d44520505a79dedfb9bc4fa3f96543e9c6.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","ca","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"},{"kind":"register_or_immediate"}]}
+    },
+    {
+      "id": "cp_async_cg_16",
+      "source_record": "int_nvvm_cp_async_cg_shared_global_16",
+      "llvm_symbol": "llvm.nvvm.cp.async.cg.shared.global.16",
+      "llvm_arguments": ["shared_ptr", "global_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f32b02e50561865fcbd41af2ab4e6d8c12485b436b14d191cbf5bbf0d248ea82."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f72afd215027bcf67444625672135ba3fbbfadbee3e28f8e90ced82b0321419e."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e091384fbc0457e593996514ad9eeafc04773c58054a8bc85a8c96980c139e5.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","cg","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"}]}
+    },
+    {
+      "id": "cp_async_cg_zfill_16",
+      "source_record": "int_nvvm_cp_async_cg_shared_global_16_s",
+      "llvm_symbol": "llvm.nvvm.cp.async.cg.shared.global.16.s",
+      "llvm_arguments": ["shared_ptr", "global_ptr", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "ptr addrspace(1)", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ce9e52734dd2438c5fc127548069d9cb7130610dc9e0752f6598268db9a3a8fe."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5081c35375c4d19aaa528f8abbeb82c57237e9596018318466d212559d1dd65b."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a0e1e9bc0608c3d6581f78c2e17d4ef3e71edcd3570f4bde0d9e8f545c8abd6.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","cg","shared","global"],"operands":[{"kind":"address"},{"kind":"address"},{"kind":"exact","value":"16"},{"kind":"register_or_immediate"}]}
+    },
+    {
+      "id": "cp_async_commit_group",
+      "source_record": "int_nvvm_cp_async_commit_group",
+      "llvm_symbol": "llvm.nvvm.cp.async.commit.group",
+      "llvm_arguments": [],
+      "llvm_results": [],
+      "concrete_llvm_arguments": [],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ea4574621afbb04281bfa8ab26db49b9a256ea5a991c63ff2907c9a725aef5a."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b2f1831e64e9e04bbe61a19f62e926eebb9e8c7a2e6e7f3fc073d97363c81f2a."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 533d9b5b9a9c927d1fe0b23c76106771cdac35407624486b921bcf4c7096811a.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","commit_group"],"operands":[]}
+    },
+    {
+      "id": "cp_async_wait_all",
+      "source_record": "int_nvvm_cp_async_wait_all",
+      "llvm_symbol": "llvm.nvvm.cp.async.wait.all",
+      "llvm_arguments": [],
+      "llvm_results": [],
+      "concrete_llvm_arguments": [],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1b3dc9f7a7dd7ebd28ad446ba1b8a97a30ffa6065ab3b9bdbb2bcb4377287859."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2dfc842c70441fe57d5a892d34441cf37fedab3955f511c5ef23118632e06924."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 96ab3b4af8cae0f8d415560129805aec75b9e3e8c26b916cda16a42cfc11d581.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","wait_all"],"operands":[]}
+    },
+    {
+      "id": "cp_async_wait_group",
+      "source_record": "int_nvvm_cp_async_wait_group",
+      "llvm_symbol": "llvm.nvvm.cp.async.wait.group",
+      "llvm_arguments": ["i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {"targets":["sm_80"],"representation":"separate typed LLVM declaration probes","stage":"declaration_canonicalization","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c4b4d445968d5e8db78f64d6903e5dab71d7e0a7ed2edd66bff7dd5443900376."},
+        {"targets":["sm_80","ptx70"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0604968fa8260d2a46de521f458b9ea4da639127e0671340e531eb5af1e5bea6."},
+        {"targets":["sm_80","ptx70"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"typed_nvvm","outcome":"succeeded","detail":"Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 77ca31b604849a0fbb4a6eeff8a57d146053d576721be6dc2212ec7347f0f1c1.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic":"cp","modifiers":["async","wait_group"],"operands":[{"kind":"immediate"}]}
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-debug-control.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-debug-control.json
similarity index 56%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-debug-control.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-debug-control.json
index a48f688eca..0e8fda8724 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-debug-control.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-debug-control.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-debug-control",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-debug-control",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_20",
@@ -19,23 +19,23 @@
       "stages": [
         {
           "targets": ["sm_20", "ptx32"],
-          "representation": "one LLVM IR probe with side-effect inline PTX for trap, brkpt, and pmevent 15",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5."
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "1,011-byte LLVM-produced PTX module containing trap, brkpt, and pmevent 15",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-elect.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-elect.json
new file mode 100644
index 0000000000..ec8133b932
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-elect.json
@@ -0,0 +1,77 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-elect",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {
+    "llvm_arguments": ["i32"],
+    "llvm_results": ["i32", "i1"],
+    "concrete_llvm_arguments": ["i32"],
+    "concrete_llvm_results": ["i32", "i1"],
+    "target_triple": "nvptx64-nvidia-cuda",
+    "gpu_target": "sm_90",
+    "ptx_feature": "+ptx80",
+    "status": "validated",
+    "declaration_attributes_canonicalized": true,
+    "runtime_validation": "unexecuted"
+  },
+  "fixtures": [
+    {
+      "id": "elect_sync_typed_probe",
+      "coverage_count": 1,
+      "stages": [
+        {
+          "targets": ["sm_90"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d6be8db0355a531bcc20af99d6b2ab9a8015c92ce61965522dec78614fd03d1c."
+        },
+        {
+          "targets": ["sm_90", "ptx80"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 de0e95f3672e5c96d15e4fd737d9225940692889c495211037f7b4f99d3cddb4."
+        },
+        {
+          "targets": ["sm_90", "ptx80"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 192c7ae696bf472409f022926d3c74f98784499ab59d769d1565e6605976c7d0.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    }
+  ],
+  "matrices": [
+    {
+      "axes": [{"name": "id", "values": ["elect_sync"]}],
+      "product_count": 1,
+      "fixtures": ["elect_sync_typed_probe"],
+      "template": {
+        "id": "${id}",
+        "source_record": "int_nvvm_elect_sync",
+        "llvm_symbol": "llvm.nvvm.elect.sync",
+        "expected_ptx": {
+          "mnemonic": "elect",
+          "modifiers": ["sync"],
+          "operands": [
+            {"kind": "register_predicate_pair"},
+            {"kind": "register_or_immediate"}
+          ]
+        },
+        "facts": {}
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-execution-control.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-execution-control.json
new file mode 100644
index 0000000000..8d133c0a33
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-execution-control.json
@@ -0,0 +1,422 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-execution-control",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "barrier_cta_sync",
+      "source_record": "int_nvvm_barrier_cta_sync_count",
+      "llvm_symbol": "llvm.nvvm.barrier.cta.sync.count",
+      "llvm_arguments": ["i32", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_30",
+      "ptx_feature": "+ptx60",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_30"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1ec32a6b202b53ac34f270d293fd60d90d5d6de4dacef2b7d7912ab16698c69f."
+        },
+        {
+          "targets": ["sm_30", "ptx60"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa2ed67b2503e249a9ae63ef2a7436c41d098c27956fe0faf441c0e961db4151."
+        },
+        {
+          "targets": ["sm_75", "ptx60"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a9e71fa97f369891143a3c32fa8c80f1c28be09176445e1edfff81c54615ffd. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "barrier",
+        "modifiers": ["sync"],
+        "operands": [
+          {"kind": "register_or_immediate"},
+          {"kind": "register_or_immediate"}
+        ]
+      }
+    },
+    {
+      "id": "barrier_cta_sync_aligned",
+      "source_record": "int_nvvm_barrier_cta_sync_aligned_count",
+      "llvm_symbol": "llvm.nvvm.barrier.cta.sync.aligned.count",
+      "llvm_arguments": ["i32", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_20",
+      "ptx_feature": "+ptx32",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_20"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c313db28069767d98c8d477dd5512e1d4dae29dc52ea0d691fe03edb03c5e50."
+        },
+        {
+          "targets": ["sm_20", "ptx32"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0615f66795bc5a379d6cf2426d83a39d4baae46194d61b6cde902e9b8c8ab367."
+        },
+        {
+          "targets": ["sm_75", "ptx32"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 999e27bbd3d4ccd52c8f1366731b70ab35d9d67ed19125ffb66ac7bf160b4bbc. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "bar",
+        "modifiers": ["sync"],
+        "operands": [
+          {"kind": "register_or_immediate"},
+          {"kind": "register_or_immediate"}
+        ]
+      }
+    },
+    {
+      "id": "barrier_cta_arrive",
+      "source_record": "int_nvvm_barrier_cta_arrive_count",
+      "llvm_symbol": "llvm.nvvm.barrier.cta.arrive.count",
+      "llvm_arguments": ["i32", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_30",
+      "ptx_feature": "+ptx60",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_30"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 70c603a8e3092012e74c6e6f5a2330489176c4af38499c379753f095ae899a2a."
+        },
+        {
+          "targets": ["sm_30", "ptx60"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6a153e4c537c56cc43402bcf7786c6540c696ff807919d11973d650c7bc9a78c."
+        },
+        {
+          "targets": ["sm_75", "ptx60"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9499277c4334862a892ac0c79eb49c79fc5c0dc889639a4b22dd5d0ea2134d0c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "barrier",
+        "modifiers": ["arrive"],
+        "operands": [
+          {"kind": "register_or_immediate"},
+          {"kind": "register_or_immediate"}
+        ]
+      }
+    },
+    {
+      "id": "barrier_cta_arrive_aligned",
+      "source_record": "int_nvvm_barrier_cta_arrive_aligned_count",
+      "llvm_symbol": "llvm.nvvm.barrier.cta.arrive.aligned.count",
+      "llvm_arguments": ["i32", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_20",
+      "ptx_feature": "+ptx32",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_20"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 171894b0dd863526a8ff4cf86ecc05f04a47032b49254ae9d5a03f58bcc5eaf4."
+        },
+        {
+          "targets": ["sm_20", "ptx32"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cb96f04a02a2ae0cde2c1d3691ca3d2752577a6b32337b39ab1afecaa7f98249."
+        },
+        {
+          "targets": ["sm_75", "ptx32"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11118e20cc0d2ec90152a19b6e75a9dbcc4f09a1d5698c030ebf890c88095fa1. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "bar",
+        "modifiers": ["arrive"],
+        "operands": [
+          {"kind": "register_or_immediate"},
+          {"kind": "register_or_immediate"}
+        ]
+      }
+    },
+    {
+      "id": "grid_dependency_launch_dependents",
+      "source_record": "int_nvvm_griddepcontrol_launch_dependents",
+      "llvm_symbol": "llvm.nvvm.griddepcontrol.launch.dependents",
+      "llvm_arguments": [],
+      "llvm_results": [],
+      "concrete_llvm_arguments": [],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_90",
+      "ptx_feature": "+ptx78",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_90"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b0abd382879e45371de563a83219bb7d4d07b5f9f42e1be22a6f4d677203678."
+        },
+        {
+          "targets": ["sm_90", "ptx78"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e1f0e1a7641d1458f0fa9d84525e830a2dbb84f58fab281009a922c356df6eec."
+        },
+        {
+          "targets": ["sm_90", "ptx78"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 306ce9690dd2d976940425ec7052d1f0cdb3b0cee03e8507345d328dde684796.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "griddepcontrol",
+        "modifiers": ["launch_dependents"],
+        "operands": []
+      }
+    },
+    {
+      "id": "grid_dependency_wait",
+      "source_record": "int_nvvm_griddepcontrol_wait",
+      "llvm_symbol": "llvm.nvvm.griddepcontrol.wait",
+      "llvm_arguments": [],
+      "llvm_results": [],
+      "concrete_llvm_arguments": [],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_90",
+      "ptx_feature": "+ptx78",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_90"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e63a89f7a0e5ef9325815f030775cc2f52a995704e695e7f6f6bc0a9d29e7e2."
+        },
+        {
+          "targets": ["sm_90", "ptx78"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 351547f724188c103868fd7cf963ba564b216c62db91f2a2a24a8685feb892a6."
+        },
+        {
+          "targets": ["sm_90", "ptx78"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 10bee9dcaa765674caa7cfcc9059aa263e5cf7b95fcb8fbdca3bee1f03e2b589.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "griddepcontrol",
+        "modifiers": ["wait"],
+        "operands": []
+      }
+    },
+    {
+      "id": "setmaxnreg_inc",
+      "source_record": "int_nvvm_setmaxnreg_inc_sync_aligned_u32",
+      "llvm_symbol": "llvm.nvvm.setmaxnreg.inc.sync.aligned.u32",
+      "llvm_arguments": ["i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_90a",
+      "ptx_feature": "+ptx80",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_90a"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6648953f506cce13f94612bf6d0da7cd3e075fbc63a6afc01f2ae709fb9a415."
+        },
+        {
+          "targets": ["sm_90a", "ptx80"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4032051215c86d96bb88330e799aec22fb09ad3d232fe483e2346692108ff967."
+        },
+        {
+          "targets": ["sm_90a", "ptx80"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06608466abf4fe979cff6b7e5e3a16a92d0daf5df071239d28e943cd7be29916.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "setmaxnreg",
+        "modifiers": ["inc", "sync", "aligned", "u32"],
+        "operands": [{"kind": "immediate"}]
+      }
+    },
+    {
+      "id": "setmaxnreg_dec",
+      "source_record": "int_nvvm_setmaxnreg_dec_sync_aligned_u32",
+      "llvm_symbol": "llvm.nvvm.setmaxnreg.dec.sync.aligned.u32",
+      "llvm_arguments": ["i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_90a",
+      "ptx_feature": "+ptx80",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_90a"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be9be58f6dcddd987a8632ca75a2ed6a592f6a65ceff960adc24bcb52bb11db3."
+        },
+        {
+          "targets": ["sm_90a", "ptx80"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddfec7d31a47990bd677ada57d94afc3bf77325e79e886219aac161fffc0f896."
+        },
+        {
+          "targets": ["sm_90a", "ptx80"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 37ad4f1284701ac5101858b0792ed16e833420d4e21cbc1c02896d03019a63a2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "setmaxnreg",
+        "modifiers": ["dec", "sync", "aligned", "u32"],
+        "operands": [{"kind": "immediate"}]
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-extended-minmax.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-extended-minmax.json
similarity index 86%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-extended-minmax.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-extended-minmax.json
index b4f8df28c5..1a3482209e 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-extended-minmax.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-extended-minmax.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "status": "validated",
@@ -21,63 +21,63 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "8 imported typed LLVM declarations and calls",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes."
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "8 typed llvm.nvvm min/max calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c)."
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed LLVM PTX 7.0 at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "8 pure production inline-PTX min/max calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match."
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "production inline PTX 7.0 at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593.",
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -90,63 +90,63 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "12 imported typed LLVM declarations and calls (every f16 modifier combination and the bf16 combinations without ftz)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb)."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed LLVM PTX 7.0 at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "production inline PTX 7.0 at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -159,63 +159,63 @@
             "sm_86",
             "ptx72"
           ],
-          "representation": "12 imported typed LLVM declarations and calls (the xorsign.abs forms)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1."
         },
         {
           "targets": [
             "sm_86",
             "ptx72"
           ],
-          "representation": "12 typed llvm.nvvm min/max calls on 16-bit floats",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40)."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa."
         },
         {
           "targets": [
             "sm_86",
             "ptx72"
           ],
-          "representation": "typed LLVM PTX 7.2 at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_86",
             "ptx72"
           ],
-          "representation": "12 pure production inline-PTX min/max calls with =h,h,h constraints",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474."
         },
         {
           "targets": [
             "sm_86",
             "ptx72"
           ],
-          "representation": "production inline PTX 7.2 at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -228,63 +228,63 @@
             "sm_86",
             "ptx78"
           ],
-          "representation": "20 imported typed LLVM declarations and calls",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes."
+          "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f."
         },
         {
           "targets": [
             "sm_86",
             "ptx78"
           ],
-          "representation": "20 typed llvm.nvvm xorsign.abs calls using LLVM's +ptx78 carrier",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a)."
+          "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec."
         },
         {
           "targets": [
             "sm_86",
             "ptx78"
           ],
-          "representation": "typed LLVM PTX 7.8 carrier",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df).",
+          "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_86",
             "ptx72"
           ],
-          "representation": "20 pure production inline-PTX calls at catalog floor PTX 7.2, mapped to LLVM +ptx78",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match."
+          "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b."
         },
         {
           "targets": [
             "sm_86",
             "ptx72"
           ],
-          "representation": "production inline PTX rewritten only to its exact PTX 7.2 admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd.",
+          "detail": "Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-f32x2.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-f32x2.json
similarity index 80%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-f32x2.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-f32x2.json
index 4103feabbe..3fd8bf1a0a 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-f32x2.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-f32x2.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [],
     "llvm_results": [],
@@ -23,23 +23,23 @@
       "stages": [
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "eight per-operation pure inline-PTX modules using one u64 carrier per f32x2 value",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d."
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1."
         },
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "eight LLVM-produced PTX 8.6 modules assembled independently for sm_100",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf.",
-          "tool_path": "/usr/local/cuda-12.9/bin/ptxas",
-          "tool_version": "CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0",
-          "tool_sha256": "983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461"
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-integer-minmax.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-integer-minmax.json
similarity index 55%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-integer-minmax.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-integer-minmax.json
index f4d3ce7c68..452c384f3f 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-integer-minmax.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-integer-minmax.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "min_relu_s32",
@@ -22,11 +22,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX min.relu.s32 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 676-byte PTX module has SHA-256 e835563e9676e88090632ce5c529840da1603ff6e65b3b88f2d722c626bd30d8. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05737e13d7ecd4d7d29377123087680401a96d0f6423842f9eb0c570c819a156."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -65,11 +65,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX max.relu.s32 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 676-byte PTX module has SHA-256 34e11547bce5d21764f671e36d47c30191aa492a57f253068f1a6f9ccb47d4cc. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 47ca916b21f4f81062a57a273ed3a2574d88dc7961f3ed061ee81135c02019a6."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -108,11 +108,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX min.s16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 2916685ee0644fd8029821bab02c8cb4155b0ed1f3cce20b10c9dbe58edacf0a. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2b9b8cd02f5dbb4b21bbe79e50ef96e87c4ff8a61d113774a049c1df97e2d645."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -150,11 +150,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX max.s16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 3a29f56af95dac7f78fa65c80cda474d1b2f71846240c18e28d1b227bed13bbe. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 019612a6c0a895437ac9e29439ca4404a44adf617072fb390270e41680557112."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -192,11 +192,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX min.u16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 1d08c966ac1bea96a509823b5dc389325f2ac8a059d04619a8ff3e428e4f5d45. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bc038f0362d6863309339980bbb161649c4774ba6fc3e43058645c320efd1724."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -234,11 +234,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX max.u16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 af376fac111934ef1cc375e57377b0eef4fd31e226db9eed52811fe89f924953. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 493b5e74e07a0232486d2df12ed0c80ed712745b36b51c17cfec678c28fdcb97."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -276,11 +276,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX min.relu.s16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 692-byte PTX module has SHA-256 c10411ef8355e759ea266343683a7f30e971822dcdf3f47eb91c74bc44a92c95. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ccd77090590ed5077784a28895d4db44c0c3601913263b77b3afedf890a94db1."
         }
       ],
       "runtime_validation": "unexecuted",
@@ -319,11 +319,11 @@
             "sm_90",
             "ptx80"
           ],
-          "representation": "pure inline PTX max.relu.s16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 692-byte PTX module has SHA-256 8d5d0d27786cd1453843d71929ee823f7f864cf2bb31388b3f35924e5a78239b. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status \"lowered\"."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5fc39021e5a2dd08e1b4c600bd65bfbd719061fd7182b4f87325793bcfeca3a."
         }
       ],
       "runtime_validation": "unexecuted",
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-basic.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-basic.json
new file mode 100644
index 0000000000..4432ef3c46
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-basic.json
@@ -0,0 +1,258 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "mbarrier_init",
+      "source_record": "int_nvvm_mbarrier_init_shared",
+      "llvm_symbol": "llvm.nvvm.mbarrier.init.shared",
+      "llvm_arguments": ["shared_ptr", "i32"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "i32"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28fcac92a1ebd5fca5408ed34c7aa64ba3546bbe5714355a7ebfd14b92f760e3."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 50ba004fcfb85da96f6f3347966e4704dda4443ace6b57676924bd528c08bf0b."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ab45efc26f5f14c513e6ab0ed925ffddbdfc1cb0e890ef4f445a40e88acebc9.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mbarrier",
+        "modifiers": ["init", "shared", "b64"],
+        "operands": [
+          { "kind": "address" },
+          { "kind": "register_or_immediate" }
+        ]
+      }
+    },
+    {
+      "id": "mbarrier_arrive",
+      "source_record": "int_nvvm_mbarrier_arrive_shared",
+      "llvm_symbol": "llvm.nvvm.mbarrier.arrive.shared",
+      "llvm_arguments": ["shared_ptr"],
+      "llvm_results": ["i64"],
+      "concrete_llvm_arguments": ["ptr addrspace(3)"],
+      "concrete_llvm_results": ["i64"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7df70d42d790f340f00c6f29da571b9b809bee9f1d093b742ce1a0e421272f46."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d140364a2d31493758ea19f5e725bbfea20f14d6ce43489b079bb4789e428ce."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57bdb80f49dc0b09308813d01303ede4a2ba34e99e63aaecd6bea6813137fd7e.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mbarrier",
+        "modifiers": ["arrive", "shared", "b64"],
+        "operands": [
+          { "kind": "register" },
+          { "kind": "address" }
+        ]
+      }
+    },
+    {
+      "id": "mbarrier_arrive_no_complete",
+      "source_record": "int_nvvm_mbarrier_arrive_noComplete_shared",
+      "llvm_symbol": "llvm.nvvm.mbarrier.arrive.noComplete.shared",
+      "llvm_arguments": ["shared_ptr", "i32"],
+      "llvm_results": ["i64"],
+      "concrete_llvm_arguments": ["ptr addrspace(3)", "i32"],
+      "concrete_llvm_results": ["i64"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3eb1f951460f6bc67c6055899542860e1010e5db5406f5e2f9accee3cbc170c2."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 174b9c4a382274b14d72edd3fd22bdc8d2486f5cb87d6c8bb4a48e7bc66b7000."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b579f9834ca9bfb229788f02fc0ce145ba59fac61cbb61036bc8c552cb4bbb7a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mbarrier",
+        "modifiers": ["arrive", "noComplete", "shared", "b64"],
+        "operands": [
+          { "kind": "register" },
+          { "kind": "address" },
+          { "kind": "register" }
+        ]
+      }
+    },
+    {
+      "id": "mbarrier_test_wait",
+      "source_record": "int_nvvm_mbarrier_test_wait_shared",
+      "llvm_symbol": "llvm.nvvm.mbarrier.test.wait.shared",
+      "llvm_arguments": ["shared_ptr", "i64"],
+      "llvm_results": ["i1"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6a81cac5ab0bf8f94c54775467e1df856112351608671f11a4e8e63ffb75f00."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df51ed0e79da4ce10d9298aa7e4fc3db4b75355044d51f95f92694acde7f9e50.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mbarrier",
+        "modifiers": ["test_wait", "shared", "b64"],
+        "operands": [
+          { "kind": "register" },
+          { "kind": "address" },
+          { "kind": "register" }
+        ]
+      }
+    },
+    {
+      "id": "mbarrier_inval",
+      "source_record": "int_nvvm_mbarrier_inval_shared",
+      "llvm_symbol": "llvm.nvvm.mbarrier.inval.shared",
+      "llvm_arguments": ["shared_ptr"],
+      "llvm_results": [],
+      "concrete_llvm_arguments": ["ptr addrspace(3)"],
+      "concrete_llvm_results": [],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx70",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76483bcaa272bf6c2c1688ff61ce630eba2cbc651af55d047b7048e43352adb7."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4effb270bc9d6ccd5d7c24a4518d532eb913541a234db702e0d50b7c4f7f0774."
+        },
+        {
+          "targets": ["sm_80", "ptx70"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 453f943cafeb60e5761ed5f0867ebb1e2519a3dd314eb9b0dc4b5e6ff6fbe65c.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mbarrier",
+        "modifiers": ["inval", "shared", "b64"],
+        "operands": [{ "kind": "address" }]
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-mbarrier-extended.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-extended.json
similarity index 56%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-mbarrier-extended.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-extended.json
index c774aee612..9ce31f4a53 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-mbarrier-extended.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-mbarrier-extended.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "status": "validated",
@@ -15,32 +15,32 @@
       "id": "inline_ptx_sm90_ptx78",
       "coverage_count": 2,
       "stages": [
-        { "targets": ["sm_90", "ptx78"], "representation": "generated extended-mbarrier inline-PTX probes", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The pinned Rust LLVM retained each exact instruction and predicate adapter at the route floor." },
-        { "targets": ["sm_90", "ptx78"], "representation": "LLVM-produced PTX 7.8 modules", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb" }
+        { "targets": ["sm_90", "ptx78"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5dbfe0da451289673522a20ec37fd0664d5276bf7760b0d8a8484d613e82ca01." },
+        { "targets": ["sm_90", "ptx78"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24bb1e758b565ffc3df774b43f1cdc6f7b26a5f827d1add384e6ffcd78747080.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc" }
       ]
     },
     {
       "id": "inline_ptx_sm90_ptx80",
       "coverage_count": 5,
       "stages": [
-        { "targets": ["sm_90", "ptx80"], "representation": "generated extended-mbarrier inline-PTX probes", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor." },
-        { "targets": ["sm_90", "ptx80"], "representation": "LLVM-produced PTX 8.0 modules", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb" }
+        { "targets": ["sm_90", "ptx80"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c." },
+        { "targets": ["sm_90", "ptx80"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc" }
       ]
     },
     {
       "id": "inline_ptx_sm90_ptx86",
       "coverage_count": 3,
       "stages": [
-        { "targets": ["sm_90", "ptx86"], "representation": "generated cluster-scope inline-PTX probes", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor." },
-        { "targets": ["sm_90", "ptx86"], "representation": "LLVM-produced PTX 8.6 modules", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas accepted each generated module for sm_90.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb" }
+        { "targets": ["sm_90", "ptx86"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97." },
+        { "targets": ["sm_90", "ptx86"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc" }
       ]
     },
     {
       "id": "nanosleep_sm70_ptx63",
       "coverage_count": 1,
       "stages": [
-        { "targets": ["sm_70", "ptx63"], "representation": "generated nanosleep inline-PTX probe", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "The pinned Rust LLVM retained exact nanosleep.u32 at the route floor." },
-        { "targets": ["sm_75", "ptx63"], "representation": "the LLVM-produced PTX 6.3 module retargeted to the nearest CUDA 13.3 assembler target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "CUDA 13.3 ptxas no longer accepts sm_70; it accepted the otherwise unchanged PTX 6.3 module for forward-compatible sm_75.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0", "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb" }
+        { "targets": ["sm_70", "ptx63"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb1f49e5307d0e617ee6cf95da7ec7a582167ac87e14e8bc4d24cb4185ee7f68." },
+        { "targets": ["sm_75", "ptx63"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 68ce336a8ad5452c31fbff9377bccbcb5540d81a83b904ec1cf2641f607fb34a.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc" }
       ]
     }
   ],
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8-f16x2.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8-f16x2.json
new file mode 100644
index 0000000000..c3cb254749
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8-f16x2.json
@@ -0,0 +1,850 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "cvt_rn_satfinite_e4m3x2_f16x2",
+      "source_record": "int_nvvm_f16x2_to_e4m3x2_rn",
+      "llvm_symbol": "llvm.nvvm.f16x2.to.e4m3x2.rn",
+      "llvm_arguments": [
+        "v2f16"
+      ],
+      "llvm_results": [
+        "i16"
+      ],
+      "concrete_llvm_arguments": [
+        "v2f16"
+      ],
+      "concrete_llvm_results": [
+        "i16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 777bd7a5362a935cb9b45c99eb5c7912d8296fd322f232be5feb4dcb71afe958."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8843a1e5b9fee11892601ac9688d35a5195bdd8c88122bb6ab5b4fa21cf76e29."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d344820320e9db759a6db9eebd370929159ce71d483a76cb70f846d27abf426."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "satfinite",
+          "e4m3x2",
+          "f16x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_satfinite_relu_e4m3x2_f16x2",
+      "source_record": "int_nvvm_f16x2_to_e4m3x2_rn_relu",
+      "llvm_symbol": "llvm.nvvm.f16x2.to.e4m3x2.rn.relu",
+      "llvm_arguments": [
+        "v2f16"
+      ],
+      "llvm_results": [
+        "i16"
+      ],
+      "concrete_llvm_arguments": [
+        "v2f16"
+      ],
+      "concrete_llvm_results": [
+        "i16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a17f8309a9b1436ceae14d34278cbc25993c1607850ec87962d7fef9fd6c3215."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11284b3464e72006408b832095a07a2ee8683336b367d675109a20b5289db239."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 82c1a28789b05b8df6d92cc52932afdc99dc761cb2bcc77c7d85e2b3a1da1e24."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "satfinite",
+          "relu",
+          "e4m3x2",
+          "f16x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_satfinite_e5m2x2_f16x2",
+      "source_record": "int_nvvm_f16x2_to_e5m2x2_rn",
+      "llvm_symbol": "llvm.nvvm.f16x2.to.e5m2x2.rn",
+      "llvm_arguments": [
+        "v2f16"
+      ],
+      "llvm_results": [
+        "i16"
+      ],
+      "concrete_llvm_arguments": [
+        "v2f16"
+      ],
+      "concrete_llvm_results": [
+        "i16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b6fcb8ea52dc533221fc3a953835ab9fb2cdb8a3c82a1dc9527dc2b980cb096."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6707ccb77a07e8b004032798bbfd07afba3af30d15568ac2c3d08d84d8b10b3."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6064f89491e6f84f9ac56c6373eec145a9b8d55507b30a4ee2ffcbe2deed7dd1."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "satfinite",
+          "e5m2x2",
+          "f16x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_satfinite_relu_e5m2x2_f16x2",
+      "source_record": "int_nvvm_f16x2_to_e5m2x2_rn_relu",
+      "llvm_symbol": "llvm.nvvm.f16x2.to.e5m2x2.rn.relu",
+      "llvm_arguments": [
+        "v2f16"
+      ],
+      "llvm_results": [
+        "i16"
+      ],
+      "concrete_llvm_arguments": [
+        "v2f16"
+      ],
+      "concrete_llvm_results": [
+        "i16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2a8ce890aff0fa2df126aba76cf0dcc3fbda114de759ad215043e6ad0d400d23."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab076ae9d2de65d0ab3caad272a118bd426726141914f2b539a30a616d655e4e."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cf21d0dd82dfc3a278f16fc60cf6772892bc2faee051ae9dd92b23f6e934f314."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "satfinite",
+          "relu",
+          "e5m2x2",
+          "f16x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_f16x2_e4m3x2",
+      "source_record": "int_nvvm_e4m3x2_to_f16x2_rn",
+      "llvm_symbol": "llvm.nvvm.e4m3x2.to.f16x2.rn",
+      "llvm_arguments": [
+        "i16"
+      ],
+      "llvm_results": [
+        "v2f16"
+      ],
+      "concrete_llvm_arguments": [
+        "i16"
+      ],
+      "concrete_llvm_results": [
+        "v2f16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b295a2577d8a238ef3098f7cb0e4fc309f9e5f1177f9f02ae766101f9d70cc67."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdceb080a053c25340804b87375d5fc8693a5fc8eaa461399d438982c1cd179a."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5682cacdfec0f3e7fe02db10b1665b3b88ab176b4db26738c965155f31ba0dbf."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "f16x2",
+          "e4m3x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_relu_f16x2_e4m3x2",
+      "source_record": "int_nvvm_e4m3x2_to_f16x2_rn_relu",
+      "llvm_symbol": "llvm.nvvm.e4m3x2.to.f16x2.rn.relu",
+      "llvm_arguments": [
+        "i16"
+      ],
+      "llvm_results": [
+        "v2f16"
+      ],
+      "concrete_llvm_arguments": [
+        "i16"
+      ],
+      "concrete_llvm_results": [
+        "v2f16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e12dd4763bbc1533bd40687021b8319bd843af8e62fc1fefe8212335310449ab."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e0264fce6d34633b3225bd3f48723a6efaa11b18a738168e34c460dfe2b16754."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a2341dac9ffd42fede2aefd66f576cac3e464c0b6b81c38773128832579eafa."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "relu",
+          "f16x2",
+          "e4m3x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_f16x2_e5m2x2",
+      "source_record": "int_nvvm_e5m2x2_to_f16x2_rn",
+      "llvm_symbol": "llvm.nvvm.e5m2x2.to.f16x2.rn",
+      "llvm_arguments": [
+        "i16"
+      ],
+      "llvm_results": [
+        "v2f16"
+      ],
+      "concrete_llvm_arguments": [
+        "i16"
+      ],
+      "concrete_llvm_results": [
+        "v2f16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a572f8ac9426c4acabd244748fbcabfffc19a7afbc3eb6b74b087a7397d0ca5."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 182d55254c86ae42a4eaaf4db2c9829f771eb0f23f6abaa92f773fd9910b607b."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7d96da22340431d9ae279f1ea4d4573bca35a171e7fc6199b5893b540a93424."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "f16x2",
+          "e5m2x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    },
+    {
+      "id": "cvt_rn_relu_f16x2_e5m2x2",
+      "source_record": "int_nvvm_e5m2x2_to_f16x2_rn_relu",
+      "llvm_symbol": "llvm.nvvm.e5m2x2.to.f16x2.rn.relu",
+      "llvm_arguments": [
+        "i16"
+      ],
+      "llvm_results": [
+        "v2f16"
+      ],
+      "concrete_llvm_arguments": [
+        "i16"
+      ],
+      "concrete_llvm_results": [
+        "v2f16"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_89",
+      "ptx_feature": "+ptx81",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b32d14ca011f505a526f0fecf85660bd93e1bbc147e728896df311e7f2b1176b."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f69dc7503144fa82436f92c37e77254ebf11ca61ffdcf492ce8b6ace6ae00234."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dff12609bfb2cbd5258c5d20e2e1c8b781090475d157dfe239cdbfc5edd9c204."
+        },
+        {
+          "targets": [
+            "sm_89",
+            "ptx81"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "declaration_attributes_canonicalized": true,
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "cvt",
+        "modifiers": [
+          "rn",
+          "relu",
+          "f16x2",
+          "e5m2x2"
+        ],
+        "operands": [
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "register"
+          }
+        ]
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8.json
new file mode 100644
index 0000000000..440df7018f
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-packed-conversion-fp8.json
@@ -0,0 +1,120 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {
+    "llvm_arguments": ["f32", "f32"],
+    "llvm_results": ["i16"],
+    "concrete_llvm_arguments": ["f32", "f32"],
+    "concrete_llvm_results": ["i16"],
+    "target_triple": "nvptx64-nvidia-cuda",
+    "gpu_target": "sm_89",
+    "ptx_feature": "+ptx81",
+    "status": "validated",
+    "declaration_attributes_canonicalized": true,
+    "runtime_validation": "unexecuted"
+  },
+  "fixtures": [
+    {
+      "id": "fp8_conversion_sm89_ptx81",
+      "coverage_count": 4,
+      "stages": [
+        {
+          "targets": ["sm_89", "ptx81"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab."
+        },
+        {
+          "targets": ["sm_89", "ptx81"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a."
+        },
+        {
+          "targets": ["sm_89", "ptx81"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": ["sm_89", "ptx81"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c."
+        },
+        {
+          "targets": ["sm_89", "ptx81"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    }
+  ],
+  "matrices": [
+    {
+      "axes": [
+        {"name": "format", "values": ["e4m3x2", "e5m2x2"]}
+      ],
+      "product_count": 2,
+      "fixtures": ["fp8_conversion_sm89_ptx81"],
+      "template": {
+        "id": "cvt_rn_satfinite_${format}_f32",
+        "source_record": "int_nvvm_ff_to_${format}_rn",
+        "llvm_symbol": "llvm.nvvm.ff.to.${format}.rn",
+        "expected_ptx": {
+          "mnemonic": "cvt",
+          "modifiers": ["rn", "satfinite", "${format}", "f32"],
+          "operands": [
+            {"kind": "register"},
+            {"kind": "register"},
+            {"kind": "register"}
+          ]
+        }
+      }
+    },
+    {
+      "axes": [
+        {"name": "format", "values": ["e4m3x2", "e5m2x2"]}
+      ],
+      "product_count": 2,
+      "fixtures": ["fp8_conversion_sm89_ptx81"],
+      "template": {
+        "id": "cvt_rn_satfinite_relu_${format}_f32",
+        "source_record": "int_nvvm_ff_to_${format}_rn_relu",
+        "llvm_symbol": "llvm.nvvm.ff.to.${format}.rn.relu",
+        "expected_ptx": {
+          "mnemonic": "cvt",
+          "modifiers": ["rn", "satfinite", "relu", "${format}", "f32"],
+          "operands": [
+            {"kind": "register"},
+            {"kind": "register"},
+            {"kind": "register"}
+          ]
+        }
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-prmt.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-prmt.json
similarity index 62%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-prmt.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-prmt.json
index 909ad2ba4c..6a9fe45fe2 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-prmt.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-prmt.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": ["i32"],
     "concrete_llvm_results": ["i32"],
@@ -22,31 +22,31 @@
       "stages": [
         {
           "targets": ["sm_20"],
-          "representation": "typed declarations for all seven PRMT records",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea."
+          "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2."
         },
         {
           "targets": ["sm_20", "ptx32"],
-          "representation": "one typed probe retaining dynamic and immediate calls for every PRMT mode",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8."
+          "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX 3.2 containing all seven PRMT modes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7.",
+          "detail": "Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-redux-f32.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-redux-f32.json
similarity index 95%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-redux-f32.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-redux-f32.json
index 993e2f6284..d00681f388 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-redux-f32.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-redux-f32.json
@@ -1,10 +1,10 @@
 {
  "schema": 5,
- "backend_profile": "rust-llvm-23.1.0-16696adc",
+ "backend_profile": "rust-llvm-23.1.3-aab11ee0",
  "backend_kind": "llvm_nvptx",
- "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
- "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
- "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+ "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+ "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+ "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
  "records": [
   {
    "id": "redux_sync_min_f32",
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-ampere-float.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-ampere-float.json
similarity index 67%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-ampere-float.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-ampere-float.json
index 1e58e60956..6f4e467a57 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-ampere-float.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-ampere-float.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_80",
@@ -20,31 +20,31 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "five typed LLVM declarations and duplicate-call probes",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691."
+          "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "five convergent side-effecting register-only inline PTX calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence."
+          "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "aggregate inline-PTX MMA fixture at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29.",
+          "detail": "Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
@@ -74,8 +74,8 @@
               "representation": "m16n8k4 TF32 lower-hardware boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the form below sm_80; the diagnostic SHA-256 is 73ef62d63bacc273dbfc152d2cb3acb93bfe3c744b3003e49cf66cd8a47e0b5d."
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the form below sm_80; the diagnostic SHA-256 is 73ef62d63bacc273dbfc152d2cb3acb93bfe3c744b3003e49cf66cd8a47e0b5d."
             }
           ]
         }
@@ -126,8 +126,8 @@
               "representation": "m16n8k8 BF16 imported-predicate boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "LLVM selects this form at its imported sm_75/PTX 6.5 predicate, but ptxas rejects it even at PTX 7.0 because this BF16 m16n8k8 form requires sm_80. The PTX 7.0 diagnostic SHA-256 is 69f7da673aef781e3eb0c06d602ccab1fdc5d95e4b0d645d6e118fd55d0747c8."
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM selects this form at its imported sm_75/PTX 6.5 predicate, but ptxas rejects it even at PTX 7.0 because this BF16 m16n8k8 form requires sm_80. The PTX 7.0 diagnostic SHA-256 is 69f7da673aef781e3eb0c06d602ccab1fdc5d95e4b0d645d6e118fd55d0747c8."
             }
           ]
         }
@@ -178,8 +178,8 @@
               "representation": "m16n8k16 F16 lower-hardware boundary",
               "stage": "ptx_assembly",
               "mechanism": "inline_ptx",
-              "outcome": "failed",
-              "detail": "ptxas rejected the form below sm_80; the diagnostic SHA-256 is f6582289bf7df33af5bf1c016ce0f73335a4a37ca257c42d201fe9cfefbe134c."
+              "outcome": "not_revalidated",
+              "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the form below sm_80; the diagnostic SHA-256 is f6582289bf7df33af5bf1c016ce0f73335a4a37ca257c42d201fe9cfefbe134c."
             }
           ]
         }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-b1.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-b1.json
similarity index 53%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-b1.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-b1.json
index 4e0d58e4b6..e0b4de6350 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-b1.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-b1.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_m8n8k128_s32_b1_xor_popc",
@@ -19,22 +19,22 @@
       "stages": [
         {
           "targets": ["sm_75", "ptx70"],
-          "representation": "convergent register-only inline PTX for m8n8k128 B1 XOR-popcount MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 666182ec743c9ffe0422975ab4ab4faf2e3ddcff656913f34303f15200fa3700."
         },
         {
           "targets": ["sm_75", "ptx70"],
-          "representation": "exact-floor PTX for m8n8k128 B1 XOR-popcount MMA",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_75 floors.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 452782b4303bcdb3357c76efeae34f913d781f5673cd5a97eace443e1dc28b77.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -62,22 +62,22 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "convergent register-only inline PTX for m16n8k128 B1 XOR-popcount MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4fc3a76fa4f4f3118a8078863963902f0b33ebeecb269fccefaf2d785a9171f4."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "exact-floor PTX for m16n8k128 B1 XOR-popcount MMA",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_80 floors.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 66efb066dd43977add4c7b6812754fc0affb5166a6dee22b060759d4eac2e5ab.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -105,22 +105,22 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "convergent register-only inline PTX for m16n8k256 B1 XOR-popcount MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e760db526d0a3aef6cdfb3a2032edb8ef925f23d7c046b15eb455e304c0a66b2."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "exact-floor PTX for m16n8k256 B1 XOR-popcount MMA",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_80 floors.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f577f397dc588d57cf9709716a198f6b17f1d4f6fa071c808bcaff2ddb58b8dc.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -148,22 +148,22 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx71"],
-          "representation": "convergent register-only inline PTX for m8n8k128 B1 AND-popcount MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6cf34f19543565ab3b0a268c1b38020771261bc358df589709391ec5e99ad3f."
         },
         {
           "targets": ["sm_80", "ptx71"],
-          "representation": "exact-floor PTX for m8n8k128 B1 AND-popcount MMA",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93107316bf0cf5ef84b40e52ec5ef683b2781a3e3a0e75737e4892184d0c6457.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -191,22 +191,22 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx71"],
-          "representation": "convergent register-only inline PTX for m16n8k128 B1 AND-popcount MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0786eb2ac48a149064adcb908b142249e47b70b7dc6b77221c691534d54f8363."
         },
         {
           "targets": ["sm_80", "ptx71"],
-          "representation": "exact-floor PTX for m16n8k128 B1 AND-popcount MMA",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d7868cbe80c20cad2dafd27dc6429792a50fd3b4efb0da2dd55b4d836fedaef9.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -234,22 +234,22 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx71"],
-          "representation": "convergent register-only inline PTX for m16n8k256 B1 AND-popcount MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b08bb4ac9ef5d95f902982d6d6e4da9bc6e758f02a86e4f0da412b6dd22f561c."
         },
         {
           "targets": ["sm_80", "ptx71"],
-          "representation": "exact-floor PTX for m16n8k256 B1 AND-popcount MMA",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd92d37a566e4ee028ef24ef3bc175b98a0ade64ebf874e29c45a409119ee2bb.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f16.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f16.json
new file mode 100644
index 0000000000..5e0be953a3
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f16.json
@@ -0,0 +1,329 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {
+    "llvm_arguments": [
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "v2f16",
+      "v2f16"
+    ],
+    "llvm_results": [
+      "v2f16",
+      "v2f16"
+    ],
+    "target_triple": "nvptx64-nvidia-cuda",
+    "gpu_target": "sm_120a",
+    "ptx_feature": "+ptx87",
+    "status": "validated",
+    "runtime_validation": "unexecuted"
+  },
+  "fixtures": [
+    {
+      "id": "dense_f8f6f4_f16_all_targets",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3."
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e."
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de."
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_100a",
+            "ptx86"
+          ],
+          "representation": "lower-PTX and disallowed-target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_110a",
+            "ptx90"
+          ],
+          "representation": "disallowed-target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_120",
+            "ptx87"
+          ],
+          "representation": "missing exact-target suffix boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_121",
+            "ptx88"
+          ],
+          "representation": "missing exact-target suffix boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx87"
+          ],
+          "representation": "lower-PTX target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx87"
+          ],
+          "representation": "lower-PTX target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx87"
+          ],
+          "representation": "lower-PTX target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        }
+      ]
+    }
+  ],
+  "matrices": [
+    {
+      "axes": [
+        {
+          "name": "a_element",
+          "values": [
+            "e2m1",
+            "e2m3",
+            "e3m2",
+            "e4m3",
+            "e5m2"
+          ]
+        },
+        {
+          "name": "b_element",
+          "values": [
+            "e2m1",
+            "e2m3",
+            "e3m2",
+            "e4m3",
+            "e5m2"
+          ]
+        }
+      ],
+      "product_count": 25,
+      "fixtures": [
+        "dense_f8f6f4_f16_all_targets"
+      ],
+      "template": {
+        "id": "mma_m16n8k32_f16_${a_element}_${b_element}",
+        "source_record": "int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_${a_element}_${b_element}_f16",
+        "llvm_symbol": "llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.${a_element}.${b_element}.f16",
+        "expected_ptx": {
+          "mnemonic": "mma",
+          "modifiers": [
+            "sync",
+            "aligned",
+            "m16n8k32",
+            "row",
+            "col",
+            "kind::f8f6f4",
+            "f16",
+            "${a_element}",
+            "${b_element}",
+            "f16"
+          ],
+          "operands": [
+            {
+              "kind": "register_list",
+              "length": 2
+            },
+            {
+              "kind": "register_list",
+              "length": 4
+            },
+            {
+              "kind": "register_list",
+              "length": 2
+            },
+            {
+              "kind": "register_list",
+              "length": 2
+            }
+          ]
+        }
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f32.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f32.json
new file mode 100644
index 0000000000..831167c597
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-f8f6f4-f32.json
@@ -0,0 +1,333 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {
+    "llvm_arguments": [
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "f32",
+      "f32",
+      "f32",
+      "f32"
+    ],
+    "llvm_results": [
+      "f32",
+      "f32",
+      "f32",
+      "f32"
+    ],
+    "target_triple": "nvptx64-nvidia-cuda",
+    "gpu_target": "sm_120a",
+    "ptx_feature": "+ptx87",
+    "status": "validated",
+    "runtime_validation": "unexecuted"
+  },
+  "fixtures": [
+    {
+      "id": "dense_f8f6f4_f32_all_targets",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e."
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451."
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab."
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_100a",
+            "ptx86"
+          ],
+          "representation": "lower-PTX and disallowed-target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_110a",
+            "ptx90"
+          ],
+          "representation": "disallowed-target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_120",
+            "ptx87"
+          ],
+          "representation": "missing exact-target suffix boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_121",
+            "ptx88"
+          ],
+          "representation": "missing exact-target suffix boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx87"
+          ],
+          "representation": "lower-PTX target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx87"
+          ],
+          "representation": "lower-PTX target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx87"
+          ],
+          "representation": "lower-PTX target boundary",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
+          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        }
+      ]
+    }
+  ],
+  "matrices": [
+    {
+      "axes": [
+        {
+          "name": "a_element",
+          "values": [
+            "e2m1",
+            "e2m3",
+            "e3m2",
+            "e4m3",
+            "e5m2"
+          ]
+        },
+        {
+          "name": "b_element",
+          "values": [
+            "e2m1",
+            "e2m3",
+            "e3m2",
+            "e4m3",
+            "e5m2"
+          ]
+        }
+      ],
+      "product_count": 25,
+      "fixtures": [
+        "dense_f8f6f4_f32_all_targets"
+      ],
+      "template": {
+        "id": "mma_m16n8k32_f32_${a_element}_${b_element}",
+        "source_record": "int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_${a_element}_${b_element}_f32",
+        "llvm_symbol": "llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.${a_element}.${b_element}.f32",
+        "expected_ptx": {
+          "mnemonic": "mma",
+          "modifiers": [
+            "sync",
+            "aligned",
+            "m16n8k32",
+            "row",
+            "col",
+            "kind::f8f6f4",
+            "f32",
+            "${a_element}",
+            "${b_element}",
+            "f32"
+          ],
+          "operands": [
+            {
+              "kind": "register_list",
+              "length": 4
+            },
+            {
+              "kind": "register_list",
+              "length": 4
+            },
+            {
+              "kind": "register_list",
+              "length": 2
+            },
+            {
+              "kind": "register_list",
+              "length": 4
+            }
+          ]
+        }
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-fp8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-fp8.json
similarity index 53%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-fp8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-fp8.json
index 24b193848e..9da652f66e 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-fp8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-fp8.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_89",
@@ -20,39 +20,39 @@
       "stages": [
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "typed LLVM declarations and representative duplicate calls",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e."
         },
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "typed and convergent inline standard FP8 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329)."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61."
         },
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "k16 F16 safe PTX floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_89", "ptx86"],
           "representation": "k16 lower-PTX boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -62,8 +62,8 @@
           "representation": "lower-SM boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -76,39 +76,39 @@
       "stages": [
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "typed LLVM declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746."
         },
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "convergent k16 F32 inline PTX",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8."
         },
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "k16 F32 safe PTX floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_89", "ptx86"],
           "representation": "k16 F32 lower-PTX boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected k16 FP8 MMA below PTX 8.7.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -118,8 +118,8 @@
           "representation": "lower-SM boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -132,39 +132,39 @@
       "stages": [
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "typed LLVM declarations and predicate audit",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55."
         },
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "convergent k32 F16 inline PTX",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9."
         },
         {
           "targets": ["sm_89", "ptx87"],
-          "representation": "k32 F16 ptxas-safe floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_89", "ptx86"],
           "representation": "k32 F16 ptxas-safe lower boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -174,8 +174,8 @@
           "representation": "lower-SM boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -188,39 +188,39 @@
       "stages": [
         {
           "targets": ["sm_89", "ptx84"],
-          "representation": "typed LLVM declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM canonicalized all four k32/F32 declarations without convergent."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f."
         },
         {
           "targets": ["sm_89", "ptx84"],
-          "representation": "convergent k32 F32 inline PTX",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84."
         },
         {
           "targets": ["sm_89", "ptx84"],
-          "representation": "k32 F32 safe PTX floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_89", "ptx83"],
           "representation": "k32 F32 lower-PTX boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
@@ -230,8 +230,8 @@
           "representation": "lower-SM boundary",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "ptxas rejected standard FP8 MMA below sm_89.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int4-m16.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m16.json
similarity index 60%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int4-m16.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m16.json
index b24371940d..e923d776a5 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int4-m16.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m16.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_m16n8k32_s32_s4",
@@ -35,25 +35,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 s4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47aa6dd3e24383098f128a64c87c47ab4797f5e033e3fc4bb0a670d2503013a."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3aa0206bc409a773e92374c337ae29700fa54735792eeaafe1d8f87fcb9c523.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -119,25 +119,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 s4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab9e908618637e0d0699f72f1a3a80b33f56c41e57a5c802af03819ef2c5ef14."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7567be8747a7b101aa8df77028eef647926a3d0dea2f0fcd83d3ebe953a622e2.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -203,25 +203,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 u4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9cd0a0a84b94f37267419e33805b31860d5685fcd2cf4564783e28343a972748."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f2bac6e9f7ce9e6cb01846a046917c140500e7e876942100e7d3ce977cbfd90d.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -287,25 +287,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 u4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a51cb813a45b9cd0c0e9dc7038bec77a14194732f4b180ec96eecd3958d6a169."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b2c092a2e9c78fd3580523dc6cc875e8903497e790eee28cfa15324c03685a7.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -371,25 +371,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 s4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d9b3d3e50e66a2281bd4a40ade4118f0afb4d7789eef88afc79e9498ac5a3ea2."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ca1b55fd6cb49b5380574ff42ec7d55b6dbf557c08ff72873a5da7eecd90838.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -456,25 +456,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 s4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 81ae38ebf4bd8a59ff18a33203383a22cace76e383f473b31a6debf0f9ee947f."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 973c539e9cb7d01605f08ede31f1bcf9fb12155ae80857ea6da2f5bf085509f6.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -541,25 +541,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 u4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 023ff2fe53a4cd8bb534791bde0e7384b5820f1fb7eb9cb1bed3627b8bed4c7e."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fedf24465346088c564e7971f6b7c29e92fceb052adb100fafb91b5e5b3b6bee.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -626,25 +626,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k32 u4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ce1cd3a7682dc726d32cf180d1b39bdf9ac3aea8f12aa2a6c6423b7f521a15ab."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a5085e6a26e349f53eba46f4e630ecd2142786cbed096ed927acc94cc9c5d910.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -714,25 +714,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 s4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5ff5fcfbf7af82721582a788d144fe6745bd1ea783aef2ff9102a328612ce3e."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95520ee8bd9137e246a8f6b10d90bf414fd6640b6941209df067082273bab6a4.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -801,25 +801,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 s4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b6f9f1214b02bafdfe07cca12072fdbfab865362fd3b295e688a738be8d0df."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 814a279d3913788ca9cc088325b84198c97e9693aaa13778c28f0f654b834e38.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -888,25 +888,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 u4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 48af8df444ba8fcff6c181e3862bbcfabfeb9817b970ddb66ea95df74814252e."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7e02b52b0a757ec780b5e93f59ff1abdea9770a82d4f0f4d1384035db9f803b.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -975,25 +975,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 u4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c9140abf7847a4b891673bba722aaef6a63ce9520ae3764efd411de6d636208f."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3824b5037253381a849ded485ed000a3c459f68a94dcd43b809a72958bf618aa.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -1062,25 +1062,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 s4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b42070729512dc4a34d959933cb2ed579c549ca2c4a34ba53634c793c2104c5."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 052fb66d0981ed2361e6cde56d62007088995b7ba713d6da1691e0012e7f4923.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -1150,25 +1150,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 s4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6d117f7dc3771deff164ddacd83b64a80d1b9c3f33543c2323394f860296b3a."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 232ee3454e7542b49e296aafcd4c9f38235fe2141fb224ca3f329aec36d456b3.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -1238,25 +1238,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 u4-by-u4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 59332065be960c742528cb03bad200c2c5a5ad472285420dd46116d3d9da8d88."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2b6ed03aae00d63ec51d849f1382eb95b8580686067589bea8eaa9433e3d8be1.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -1326,25 +1326,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "convergent register-only inline PTX for m16n8k64 u4-by-s4 INT4 MMA",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa313e124cffc714546231b65e670c63303446e7edb95dce4af94ca57bf258a3."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ed63a17572d63f68c5f40f43cf5920d3432723ec6b946bd1689e3c44d3fe4f9.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int4-m8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m8.json
similarity index 50%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int4-m8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m8.json
index b75f3096d0..0dd1dc3a31 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int4-m8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int4-m8.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_m8n8k32_s32_s4",
@@ -18,21 +18,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k32 signed INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping s4-by-s4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 242a7d2104a977a3c344c947eb5c03d10a18711cb776b10a27c2519d3760eeb0."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c5b6ee75c034c9af33c18ab6c91f725b42d2b54c8bf132ec1b89c115cb9a3548.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -49,21 +49,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k32 mixed-sign INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping s4-by-u4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57e936cd723c8522e8d7ad7c20eec7f1efe473d9dde5381bc74874bbd737acc1."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 540c2a2b1c2c2945022dad6f87d3cdf948ade2eab7a1aa2149e47b320adf4e48.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -80,21 +80,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k32 unsigned INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping u4-by-u4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16ea31bd434789327320d0683d1cc83a11058292365655503c77c2143139e1e4."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75676b7cb4e4fb123537b9f8f1cc40ccb04e1db0271ab3eb760961f4ea73ea37.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -111,21 +111,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k32 mixed-sign INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping u4-by-s4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a777e8a890e81f7bc799e08ffa31129ff28d46f13cd9bbb9467f7d3e09e294b6."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 87501053789841d181fd1aae8dac5b4dc516639d99983f1867fd6f4df9be4836.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -142,21 +142,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k32 signed INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite s4-by-s4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5fb1110c752bf4fe08f37bd8b7f8345d676ca3eea504e71bed250fb3c7497ce."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e9c1003612b5e1b0441919a9d51eeadbd9a9c26f365be978cf5d279b0b0d1fc.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "satfinite", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -173,21 +173,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k32 mixed-sign INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite s4-by-u4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1a0ad568976eb0b009eeb8941f96e74169d76cd8f5e486c761848b16a6a78f68."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ee342c6e9f3f578c32c6af5dd1c2668cf556669258abfcce6f85298f77a2fe90.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "satfinite", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -204,21 +204,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k32 unsigned INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite u4-by-u4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b77eb2760494d673de880beb0e9da2110a690897b8968ba0b33d57e0f7b635f."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 666fe69ffd45aab6a3f5bfc5183c41460a862957228ada97aee5dece4d4b92a4.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "satfinite", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -235,21 +235,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k32 mixed-sign INT4 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite u4-by-s4 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d3a2ccb17359a5a84b5f2ffbbbf21bdbf3e862d6a2a6e007cba7442aa65144a9."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 787d75bc93a0a84ac830fa9dd7472adc96e9b78362d7e6ce1b107301d5c6d0a6.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k32", "row", "col", "satfinite", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int8-m8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8-m8.json
similarity index 50%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int8-m8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8-m8.json
index 71200f84ab..bdbb1e3738 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int8-m8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8-m8.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_m8n8k16_s32_s8",
@@ -18,21 +18,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k16 signed INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping s8-by-s8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5c2da0b56637465b6318014f3bb75f68d843c14ef827e70a7ef937625ad5e2c."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df22078cac5b20809b7d5a9eea6a29d6cfe354500d77fa360d86a0b4a1023852.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -49,21 +49,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping s8-by-u8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab1d12b93c4b140169d3169350a16500008aa4163f09614a631b7e1dc4fdbcb1."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a423b06a09e13590b1d54c6f8d8be51d1e394145354b004a635293e1d76ac83c.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -80,21 +80,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k16 unsigned INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping u8-by-u8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f6d577d18158476e7ab49d17a2ad0f47da8b6ae752a6ed080d20e1380d52bee."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1955df76beb6ee81ed7ea879c3200f4f6bb7e1966deb92e945b3f325841dc5cd.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -111,21 +111,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for m8n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping u8-by-s8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f75f6988bc1dda288d7c571d72462c4298a17b37aaa830a033de583852ac772."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4dff1215835b5b26444fd0e25f94789efea21a227da12d555c889c796c0ea93f.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -142,21 +142,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k16 signed INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite s8-by-s8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 270630238dd5c5f9cf517a25b950096e73e12e4479f81ac8c1cd3283017390d0."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1bf5493b1e849d63814126e394e6f0572529d702a2433d0de1ff47b7b4ac0650.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -173,21 +173,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite s8-by-u8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d10485a6ecf25b1ef218ba5ca3641e58f8066afb750fe46d83566b3fe2b4f072."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f5aaa3aac438dbac75e659a7a874e86e99e3f0114d38a53f3d889f2063ea0b6.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -204,21 +204,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k16 unsigned INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite u8-by-u8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51db7a6fe697ef26e445bac4abdaa4d5e45e887e8aa5a366f608654d3466aab0."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06aafc80a8ca8190b809819eaac9c4776807f6380a05ed5a02ce479898506506.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
@@ -235,21 +235,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_75", "ptx65"],
-        "representation": "convergent register-only inline PTX for saturating m8n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite u8-by-s8 instruction in C, A, B API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93bd9995092da3436c05eb1720405ceeea64728cf2eb687f907c7b1c820d0148."
       }, {
         "targets": ["sm_75", "ptx65"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67dbf19d3684c67bd2b2ffface224d08f9e24a216e617ebf5307e530bc492d97.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k16", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8.json
similarity index 51%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8.json
index 4355ca9c09..aff9f53000 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma-int8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-int8.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_m16n8k16_s32_s8",
@@ -18,21 +18,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k16 signed INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping s8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a48e36d7da44b51913001f1eb7d4d1a4c3fe187828db3a37451079cfb79bd041."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3f2318b8382d87d1f67a2cc34a6842dbb8d910a85130f0d3d991690b39d6f1c2.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -49,21 +49,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping s8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c7ad5689bb1dc7d9f6e66a3b7024246cecd620c8499b9468999c5efc7cf4ef38."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2318600ab58b33347303547db32bfbfdccea5210d1401891255d10c18a8d89cd.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -80,21 +80,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k16 unsigned INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping u8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 db2d33329053739e1f39831d56e2d27e487dd334679bb09de2f5c8f80eb1de65."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0a674a423572b924a7f2e6dd0a2c69d601e5d2c4682dfa1414103e7045a686c6.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -111,21 +111,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping u8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9921a009519c555a2adb72baf0e136405ef6b850b9a19d83f10e7d0b3d763f71."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 969755d83267f71bbb72953caff716f8eae65ee1d14c4a293879a0a21338c337.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -142,21 +142,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k32 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping s8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 77110080515c64834cb555469a0ea69ed12c499b30be821f8832fdcfac0f73c5."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7700bbce0406dc6977a0af2702a6679b1a41e88c867eed3903621b9d14beef53.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -173,21 +173,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k32 unsigned INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping u8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67ac1fcd13f430b3a14dc6d8467cc9beec1a0b175c1e785fce2a9a91545d379d."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ba09c1ddf924a125f7171df66dec143a9c4d598837bdaed8de694fc7b52b4b8.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -204,21 +204,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for m16n8k32 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact wrapping u8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ec734a83add935db593a1f90090b395a74af19c9c84661e61a80d46ba07966ca."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13d684b6400fcb97db851b1a163797e7b6cabd462f847f3f54b3342a0d49697c.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -235,21 +235,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k16 signed INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite s8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32f364dd111daf1d85405149c434903c9ba832ee2d5d6e3067ab468e61bcd713."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f136aaf2137b3c1f962fdcdbeaf7f7ed4ad688be4fb5cd55dadaf7ac100fc153.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -266,21 +266,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite s8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 516ff5a0c66113c514e59c0272876c73360c509581521ac8ef9c6c7b1a02c82e."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b15a9cd586a5fbde549415441be60eae06e57d99604ba2350ca476c0066f6a1a.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -297,21 +297,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k16 unsigned INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite u8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 332115a18b93fa8d2c7916c9fe028636dde30234d0b197244146f55eca052148."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8301b6cd488dea73b3278090a9a30910809f2afba5c3e9b58b21a1d84cdca8bd.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -328,21 +328,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k16 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite u8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab5b0e911466d87855d48864c54ff2da53c2e4e5b940b1c0812b378ea453e585."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a300eed5eea4deeaf14e77ce212087a11d0a912a110956a8f98d7ef0219018cf.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 4}]}
@@ -359,21 +359,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k32 signed INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite s8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c512bacbd7f6da60ef9eda11b065823f600b9f0409a0ca562d8b2ae0e2c2df48."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57e571ed96c84be6aaa7c9428a16c4c002f791b55429f427820fcc3d6fbb177c.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -390,21 +390,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k32 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite s8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16e7d991475a0e8577e129c8337f9eb2ba08ec792a3500cd9890a5be58311b2b."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 74351d962b7eb0ac0ca27e606b8d6e75468b39752b667b28e6277475de18bb29.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -421,21 +421,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k32 unsigned INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite u8-by-u8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eaabf735e8ebc1b813e38c3ec0d3c20f167786603c98ce8ebc1867abdf0e5d52."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 deefb4ba049125c5b91b39ac823909f05e24c773ceac2520b10210d11e02c17b.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -452,21 +452,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for saturating m16n8k32 mixed-sign INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact satfinite u8-by-s8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f2c069f3d47afb6653954d8f58fb146b9e927136d703104a92a2b8312c874d45."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fffbf71c999883fa695a7300a982460fc7e6425a322521cc9e98523dcfdc3fc2.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-mxf8f6f4-f32.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-mxf8f6f4-f32.json
new file mode 100644
index 0000000000..3358eff0fe
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma-mxf8f6f4-f32.json
@@ -0,0 +1,235 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {
+    "llvm_arguments": [
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "i32",
+      "f32",
+      "f32",
+      "f32",
+      "f32",
+      "i32",
+      "i16",
+      "i16",
+      "i32",
+      "i16",
+      "i16"
+    ],
+    "llvm_results": [
+      "f32",
+      "f32",
+      "f32",
+      "f32"
+    ],
+    "target_triple": "nvptx64-nvidia-cuda",
+    "gpu_target": "sm_120a",
+    "ptx_feature": "+ptx87",
+    "status": "validated",
+    "runtime_validation": "unexecuted"
+  },
+  "fixtures": [
+    {
+      "id": "dense_mxf8f6f4_f32_all_targets",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872."
+        },
+        {
+          "targets": [
+            "sm_120a",
+            "ptx87"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25."
+        },
+        {
+          "targets": [
+            "sm_120f",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e."
+        },
+        {
+          "targets": [
+            "sm_121a",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx88"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27."
+        },
+        {
+          "targets": [
+            "sm_121f",
+            "ptx88"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    }
+  ],
+  "matrices": [
+    {
+      "axes": [
+        {
+          "name": "a_element",
+          "values": [
+            "e2m1",
+            "e2m3",
+            "e3m2",
+            "e4m3",
+            "e5m2"
+          ]
+        },
+        {
+          "name": "b_element",
+          "values": [
+            "e2m1",
+            "e2m3",
+            "e3m2",
+            "e4m3",
+            "e5m2"
+          ]
+        }
+      ],
+      "product_count": 25,
+      "fixtures": [
+        "dense_mxf8f6f4_f32_all_targets"
+      ],
+      "template": {
+        "id": "mma_m16n8k32_mxf8f6f4_f32_${a_element}_${b_element}",
+        "source_record": "int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_${a_element}_${b_element}_f32_ue8m0",
+        "llvm_symbol": "llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.${a_element}.${b_element}.f32.ue8m0",
+        "expected_ptx": {
+          "mnemonic": "mma",
+          "modifiers": [
+            "sync",
+            "aligned",
+            "m16n8k32",
+            "row",
+            "col",
+            "kind::mxf8f6f4",
+            "block_scale",
+            "f32",
+            "${a_element}",
+            "${b_element}",
+            "f32",
+            "ue8m0"
+          ],
+          "operands": [
+            {
+              "kind": "register_list",
+              "length": 4
+            },
+            {
+              "kind": "register_list",
+              "length": 4
+            },
+            {
+              "kind": "register_list",
+              "length": 2
+            },
+            {
+              "kind": "register_list",
+              "length": 4
+            },
+            {
+              "kind": "register"
+            },
+            {
+              "kind": "register_list",
+              "length": 2
+            },
+            {
+              "kind": "register"
+            },
+            {
+              "kind": "register_list",
+              "length": 2
+            }
+          ]
+        }
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma.json
similarity index 50%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma.json
index c0b933c042..1105384f7c 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-register-mma.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-register-mma.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_m16n8k16_f32_bf16",
@@ -18,21 +18,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for BF16 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact BF16 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0a3bd41a37b501ef0419e036d4a2325f28dee3d5d47c050ab1af6d52b9904385."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b0d9dbc95cc97fb422f1b7b9e39d1d68c866b4ffcf48c9a32f8fb917454bb82.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "f32", "bf16", "bf16", "f32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -49,21 +49,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for F16 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact F16 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cb21825affcdfcb250dea8bfead8827d200376f1b6dfb69b7659d670b4fe0a8d."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d25700d38dead0336c44add15c4dc07020aab017e285b96e0e94bf091e743e4e.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k16", "row", "col", "f32", "f16", "f16", "f32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -80,21 +80,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for TF32 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact TF32 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 229d640dcb7295b409e4d690f6094a704af0b77e69e834abb5c2c9d26ac5daa0."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7543fed70ec348791089d56f1eee6d51e13501a1a40092f118bec2c99b734d3c.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k8", "row", "col", "f32", "tf32", "tf32", "f32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -111,21 +111,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for signed INT8 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact signed INT8 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6524c9c1c5d31e51a7bc4f69ed88bb6f64bc9718c4cd4e6dbdf38d71cceb7406."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f67ea6f8fc76d2142c8297e2fbef29257d5a89eba91380d4ebbbf47cdce91489.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m16n8k32", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}]}
@@ -142,21 +142,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx70"],
-        "representation": "convergent register-only inline PTX for f64 MMA",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The generated probe emitted the exact f64 MMA instruction with C, A, B operands kept in public API order."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 455ca35747a458536fae20c94d1749e6b7c07dc9b482fb469192376c9e10a1a8."
       }, {
         "targets": ["sm_80", "ptx70"],
-        "representation": "PTX emitted by the selected Rust LLVM register-MMA probe",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f7bf4efce75e805867c6ef541acdd239b2df1f6819376e1d08b24cf76398b507.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sync", "aligned", "m8n8k4", "row", "col", "f64", "f64", "f64", "f64"], "operands": [{"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 1}, {"kind": "register_list", "length": 2}]}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-arithmetic.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-arithmetic.json
similarity index 75%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-arithmetic.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-arithmetic.json
index efaf4695f3..14434e7c47 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-arithmetic.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-arithmetic.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_80",
@@ -20,31 +20,31 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "36 selected typed LLVM declarations within the 44-candidate proof bundle",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba."
+          "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "36 selected typed llvm.nvvm mul, div, and fma calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared."
+          "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "36 selected typed LLVM routes within the 44-output assembly proof",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests.",
+          "detail": "Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -54,31 +54,31 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "12 selected typed LLVM add declarations within the 20-candidate proof bundle",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "12 selected typed llvm.nvvm.add calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted."
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "12 selected typed LLVM add routes within the 20-output assembly proof",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs.",
+          "detail": "Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -88,31 +88,31 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "16 imported LLVM 22 saturating add and fma declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60."
+          "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "16 pure inline-PTX saturating add and fma calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b."
+          "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "16 inline-PTX saturating scalar instructions assembled at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7.",
+          "detail": "Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-conversion.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-conversion.json
similarity index 53%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-conversion.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-conversion.json
index 0764619e85..e05a36950d 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-conversion.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-conversion.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": ["f32"],
     "llvm_results": ["i32"],
@@ -22,31 +22,31 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "canonical typed LLVM declaration for cvt.rna.tf32.f32",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized the fresh 598-byte probe to 584-byte IR (SHA-256 2448daf9235fa4db9287c592309f7496856e9cf31929014a569c70813978d9c8) with the imported f32-to-i32 signature and canonical intrinsic attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7fc68746b0ee106bcb3270f8a02cb83cd0a4e85363c0854f82546717a33bf72d."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "typed llvm.nvvm.f2tf32.rna call",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc lowered the fresh probe to 565-byte PTX (SHA-256 19695568a4fde5c5847f25e5103f5f4aecf6b96dd9f408028fee66222fe8c82f) containing cvt.rna.tf32.f32."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5355197dbd45ff46716deac34ba4d8de1084ee4374f537af9a67dd2fb47b6c5."
         },
         {
           "targets": ["sm_80", "ptx70"],
-          "representation": "typed LLVM output assembled at the catalog floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the fresh sm_80/PTX 7.0 output and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8bf2670d9bea4e04e196ffe05f2954a5d9f8beefc57a231af094444b20348c3a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -56,31 +56,31 @@
       "stages": [
         {
           "targets": ["sm_80", "ptx81"],
-          "representation": "canonical typed LLVM declaration for cvt.rna.satfinite.tf32.f32",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized the fresh 628-byte probe to 634-byte IR (SHA-256 51e877e213ff97519d0a1c4080afee97c3e4b208b88ab2071b214a36428c77f4) with the imported f32-to-i32 signature and canonical intrinsic attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a38ab8cd30065c8554fc11f7b6f155740eb533d290525b57a45a4b339465234."
         },
         {
           "targets": ["sm_80", "ptx81"],
-          "representation": "typed llvm.nvvm.f2tf32.rna.satfinite call",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc lowered the fresh probe to 634-byte PTX (SHA-256 be00dc0e407a9f71d14c5b73a5a4c53b06e4430ad20f888e5595499b25c348ad) containing cvt.rna.satfinite.tf32.f32."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2819c3052033ee04b85149f6891347ea860e7bacc511a0e7edb1d5fb4f36cd2."
         },
         {
           "targets": ["sm_80", "ptx81"],
-          "representation": "typed LLVM output assembled at the catalog floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the fresh sm_80/PTX 8.1 output and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e8c2a92a43cdb367ea7856ca832f6fea44f45613b70f89f2acd2b45c4d8b35b4.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -90,31 +90,31 @@
       "stages": [
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "canonical typed LLVM declarations for the rn, rn.relu, rz, and rz.relu conversions",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "typed LLVM calls for the four PTX 7.8 scalar conversions",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "typed LLVM outputs assembled at the catalog floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -124,31 +124,31 @@
       "stages": [
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "canonical typed LLVM declarations for the rn/rz satfinite and relu.satfinite conversions",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71."
         },
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "typed LLVM calls for the four PTX 8.6 scalar conversions",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb."
         },
         {
           "targets": ["sm_100", "ptx86"],
-          "representation": "typed LLVM outputs assembled at the catalog floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-math.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-math.json
similarity index 82%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-math.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-math.json
index 1e0ee2c097..fb08481974 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-scalar-math.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-scalar-math.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_80",
@@ -23,37 +23,37 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "28 typed LLVM declarations for rcp and sqrt variants with DAG selection patterns",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets."
+          "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "28 typed llvm.nvvm rcp and sqrt calls lowered through DAG selection",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3."
+          "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "28 typed LLVM routes within the 37-function PTX proof",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+          "detail": "Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -66,37 +66,37 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "9 LLVM declarations for sin, cos, lg2, and rsqrt variants without DAG selection patterns",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching."
+          "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "9 LLVM-lowered unary math calls in the 37-function PTX proof",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3."
+          "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "9 inline-PTX scalar math instructions assembled at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2.",
+          "detail": "Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -109,29 +109,29 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "3 appended inline-PTX unary math probes (ex2.approx.{f32,ftz.f32}, tanh.approx.f32) attested individually",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback.",
-          "tool_path": "llc",
-          "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb.",
+          "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+          "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "3 appended inline-PTX scalar math instructions assembled at the admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor.",
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -141,23 +141,23 @@
       "stages": [
         {
           "targets": ["sm_75", "ptx70"],
-          "representation": "llvm.nvvm.ex2.approx.f16 lowered at the documented hardware and PTX floor",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc selected ex2.approx.f16 for sm_75 with +ptx70; PTX SHA-256 21a1d2fe684434a6c71b2454ff0a26463002def6ab7c09e00252a70d9770b7df. The same probe fails selection at sm_70 or +ptx65. llvm.nvvm.ex2.approx.ftz.f16 and llvm.nvvm.ex2.approx.bf16 fail selection; llvm.nvvm.ex2.approx.ftz.bf16 selects at sm_90 with +ptx78."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cee684a507daecda40b46b40a8de9a2f2582dfcf5aae88b3d844b49e614c44e7."
         },
         {
           "targets": ["sm_75", "ptx70"],
-          "representation": "single ex2.approx.f16 instruction assembled at its admission floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas assembled the 640-byte sm_75/PTX 7.0 output into a 1,464-byte cubin with SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0fd052b8dd9f1821ac995eba3175320422af7c36dcdf0fc99190a1b2819de6e6.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-fp8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-fp8.json
similarity index 59%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-fp8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-fp8.json
index cc6111b3c8..fc08f17ff1 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-fp8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-fp8.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [
       "i32",
@@ -44,25 +44,25 @@
             "sm_89",
             "ptx84"
           ],
-          "representation": "four convergent plain-sparsity sparse FP8 F32 inline-PTX probe modules in D, A, B, C, metadata, selector order",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047."
         },
         {
           "targets": [
             "sm_89",
             "ptx84"
           ],
-          "representation": "exact-floor plain-sparsity sparse FP8 F32 PTX",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms.",
-          "tool_path": "/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
           "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8-m16n8k64.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8-m16n8k64.json
new file mode 100644
index 0000000000..5dd10405f4
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8-m16n8k64.json
@@ -0,0 +1,146 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "mma_sp_m16n8k64_s32_s8",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcd1b6080f311a524c25afacc87c7db3e52597ff2da06494ecc63a319740ff3a."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7c617c1291bbecb8b6aa72f56558797c56ea20bb7a78fd5769aae5d0f12b720.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_s8_u8",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s8_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s8.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9dcb133309ca12ab7c5fabd0703c014103473203370652bf89f8b5c39de9898."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b95b07cec0fa97e908c16d96d9ded6697b9173ef6debe699b291aa890a5d22ee.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u8",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 73597b682d1b12c2fccfad62f40ece013c3c3fcd9024e59c32ef84a8732364e5."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ac63519d007e8dce250d4517871606b252765146f7f59abdd1602790cb331543.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u8_s8",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u8_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u8.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c4d93f3a3ddbc1b78370fc6d41ef8514a6a7aabaa7ec338b788f32dd08f910ae."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46b4764da009f8939b74a6235f29b9a6fed892e8b86f68ff56d576e2e5d11151.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_s8_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f1627fbb5b81a5df542173347c1dabb3e3493c579f2d1ec962fbfb8bf88936a7."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 207e717b6d247cfb221fdbcec7a626708ce5b740bdb21f409d1dbbb439941c5e.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_s8_u8_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 35ffdccb3d2495814cf57898071b7b916a6b6ca5627fa62e179523389496a18d."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 619c473b81d7936bbd80cd8b1195a5911b756c6209ce85beba9dc3b63c929df0.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u8_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a3f0063a7b3cacd62f63384b92d8719e8a4697c12df321ff9a1d5dece0e84d13."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ff140019dd06530e11bd9507f0e371f07c5d88300c8dbbcd619681437d175825.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u8_s8_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx71"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ee679b12117e5c9162717c0d88a751b96da7576d08049d06c19e3fb6c6300c73."},
+        {"targets": ["sm_80", "ptx71"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 00174476515d29940f2a3bd1efce00363d2005a85306c962ce53c6a9bc84e43b.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-int8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8.json
similarity index 52%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-int8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8.json
index b0130c20ad..2e41f3c2b5 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-int8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-int8.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_sp_m16n8k32_s32_s8",
@@ -18,21 +18,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 signed INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping s8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6117422b739d9c39ae85c36a064b6ba0cfe6242cbd6c19fffe7ff68365e00a79."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a84718234f047484ec8d86c72e2a333a2982ea25b4266f109070266baa49c715.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -49,21 +49,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping s8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32f1c01ee1a43ca81812d076ee8db48b8928aa441daf3f71dfc025f6b50a516a."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 875de648ec423e0f311608024d48bd81ae45ae442c9741a8f08c01502f863e9b.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -80,21 +80,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 unsigned INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping u8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34822f024cc535508a1d7ae0c22eb25c61a337e07e674e4c5d0821c85d2e44e0."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c0cad4768d71256792c04fdf7e962f3630275a62a909d6ecd15a67a1aacb7765.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -111,21 +111,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping u8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1dc739b2f8f0bc7f5d2a744a638f9f9e93a9c9763584bdc4ef239e3b740a7900."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11793a16a1edc4a91afc40987b925e8829ac10289034931cf23a5dd19c54ff17.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -142,21 +142,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 signed saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating s8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3c5d9bb6256f7009234918046b2dfb11507265c220722e1a42aa20653a409aec."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9c0a5d19a7e1637ff335da9a214e99e06463d5160df778e1bad3dfcfc6874890.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -173,21 +173,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating s8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b79bb4f92abfb0ce7a3b13baed56134e8043e73d4552d305d42f8f2d97661bfd."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 adbd36486601138ffe021f493473ae535351c523380e780a11678e1a4da0e4e7.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -204,21 +204,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 unsigned saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating u8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d8c7f3a382ee9d1ea29abdb323d012394df4bf6bce1fa24e8594c78e8bfa65f7."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46586ec2c81c1e15a80b74485e9e14cad94c67af27deead4a2ce433787a814af.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -235,21 +235,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx71"],
-        "representation": "convergent sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The pinned LLVM 22 combined probe retained both selector forms and the exact saturating u8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d51f3473017b1bc3eafcae85544debb79cd50a621baebf76491730e6af4dfd83."
       }, {
         "targets": ["sm_80", "ptx71"],
-        "representation": "exact-floor PTX for base sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c8b4cd0226a362e5b29976f9f58f7fba85de7a0f25858de6e77bef814660815e.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-ampere-float.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-ampere-float.json
similarity index 73%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-ampere-float.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-ampere-float.json
index df9e933659..55c39f8228 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-ampere-float.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-ampere-float.json
@@ -1,18 +1,18 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
-  "defaults": {"target_triple":"nvptx64-nvidia-cuda","gpu_target":"sm_80","ptx_feature":"+ptx85","status":"executed","runtime_validation":"executed"},
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {"target_triple":"nvptx64-nvidia-cuda","gpu_target":"sm_80","ptx_feature":"+ptx85","status":"validated","runtime_validation":"unexecuted"},
   "fixtures": [{
     "id": "ordered_ampere_float_all",
     "coverage_count": 8,
     "stages": [
-      {"targets":["sm_80","ptx85"],"representation":"eight committed per-intrinsic convergent inline-PTX probe fixtures covering every selector immediate","stage":"backend_codegen","mechanism":"inline_ptx","outcome":"succeeded","detail":"Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86."},
-      {"targets":["sm_80","ptx85"],"representation":"exact-floor per-intrinsic probe PTX containing all eight ordered sparse floating MMA forms","stage":"ptx_assembly","mechanism":"inline_ptx","outcome":"succeeded","artifact_kind":"cubin","detail":"CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0","tool_sha256":"7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"},
-      {"targets":["sm_120","ptx85"],"representation":"all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle","stage":"runtime","mechanism":"inline_ptx","outcome":"succeeded","detail":"Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors."}
+      {"targets":["sm_80","ptx85"],"representation":"separate probes for the recorded lowering mechanism","stage":"backend_codegen","mechanism":"inline_ptx","outcome":"succeeded","detail":"Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5."},
+      {"targets":["sm_80","ptx85"],"representation":"separate PTX outputs assembled for the recorded target","stage":"ptx_assembly","mechanism":"inline_ptx","outcome":"succeeded","artifact_kind":"cubin","detail":"Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2.","tool_path":"/usr/local/cuda-13.3/bin/ptxas","tool_version":"Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0","tool_sha256":"afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"},
+      {"targets":["sm_120","ptx85"],"representation":"all eight generated ordered sparse floating MMA APIs checked across all 32 lanes and four logical accumulators per lane against an independent host GEMM oracle","stage":"runtime","mechanism":"inline_ptx","outcome":"not_revalidated","detail":"Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors."}
     ]
   }],
   "matrices": [
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f16.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f16.json
new file mode 100644
index 0000000000..c46e709c04
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f16.json
@@ -0,0 +1,163 @@
+{
+  "schema": 6,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "defaults": {
+    "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "v2f16", "v2f16", "i32", "i32"],
+    "llvm_results": ["v2f16", "v2f16"],
+    "target_triple": "nvptx64-nvidia-cuda",
+    "gpu_target": "sm_120a",
+    "ptx_feature": "+ptx87",
+    "status": "validated",
+    "runtime_validation": "unexecuted"
+  },
+  "fixtures": [
+    {
+      "id": "ordered_f8f6f4_f16_sm120a_ptx87",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": ["sm_120a", "ptx87"],
+          "representation": "separate typed LLVM declaration probes",
+          "stage": "declaration_canonicalization",
+          "mechanism": "typed_nvvm",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e."
+        },
+        {
+          "targets": ["sm_120a", "ptx87"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f."
+        },
+        {
+          "targets": ["sm_120a", "ptx87"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    },
+    {
+      "id": "ordered_f8f6f4_f16_sm120f_ptx88",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": ["sm_120f", "ptx88"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a."
+        },
+        {
+          "targets": ["sm_120f", "ptx88"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    },
+    {
+      "id": "ordered_f8f6f4_f16_sm121a_ptx88",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": ["sm_121a", "ptx88"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec."
+        },
+        {
+          "targets": ["sm_121a", "ptx88"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    },
+    {
+      "id": "ordered_f8f6f4_f16_sm121f_ptx88",
+      "coverage_count": 25,
+      "stages": [
+        {
+          "targets": ["sm_121f", "ptx88"],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2."
+        },
+        {
+          "targets": ["sm_121f", "ptx88"],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ]
+    }
+  ],
+  "matrices": [
+    {
+      "axes": [
+        {"name": "a_element", "values": ["e2m1", "e2m3", "e3m2", "e4m3", "e5m2"]},
+        {"name": "b_element", "values": ["e2m1", "e2m3", "e3m2", "e4m3", "e5m2"]}
+      ],
+      "product_count": 25,
+      "fixtures": [
+        "ordered_f8f6f4_f16_sm120a_ptx87",
+        "ordered_f8f6f4_f16_sm120f_ptx88",
+        "ordered_f8f6f4_f16_sm121a_ptx88",
+        "ordered_f8f6f4_f16_sm121f_ptx88"
+      ],
+      "template": {
+        "id": "mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_${a_element}_${b_element}_f16",
+        "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_${a_element}_${b_element}_f16",
+        "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.${a_element}.${b_element}.f16",
+        "expected_ptx": {
+          "mnemonic": "mma",
+          "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "kind::f8f6f4", "f16", "${a_element}", "${b_element}", "f16"],
+          "operands": [
+            {"kind": "register_list", "length": 2},
+            {"kind": "register_list", "length": 4},
+            {"kind": "register_list", "length": 4},
+            {"kind": "register_list", "length": 2},
+            {"kind": "register"},
+            {"kind": "immediate"}
+          ]
+        }
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f32.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f32.json
similarity index 61%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f32.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f32.json
index dfefd3fe08..9946fde606 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-f8f6f4-f32.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-f8f6f4-f32.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "f32", "f32", "f32", "f32", "i32", "i32"],
     "llvm_results": ["f32", "f32", "f32", "f32"],
@@ -21,23 +21,23 @@
       "stages": [
         {
           "targets": ["sm_120a", "ptx87"],
-          "representation": "convergent ordered kind::f8f6f4 F32 inline PTX in public C, A, B, metadata, selector order",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7."
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b."
         },
         {
           "targets": ["sm_120a", "ptx87"],
-          "representation": "exact-floor ordered kind::f8f6f4 F32 PTX",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected.",
+          "detail": "Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k128.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k128.json
new file mode 100644
index 0000000000..93fa8dd5e0
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k128.json
@@ -0,0 +1,146 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7797f8b2f85799bd268bf06f71383c5e441791fb25d6c11afe368dde7b688f2a."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddf20d448ac238027732e5de25eceff5461797fae6dadac35f3ff3a32a85a89.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4_u4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f1dcbdef6c39dbf4dd3542b151d9ca227d9ba45ff1e57d7f5d5ee7a93f09a0d4."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cc679e863a7f3653d5ef77a16942b4107959b879360a5b4cf0dd0c3d237cc3f6.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 943b9548d42ec57747affb74e4de43cd2f8b6dffcc84b3f57c61c014622672e4."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46e7ba5d6bd8e53adf639c09a6f053a9d08edb608b10649a80c737ced39afaa4.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4_s4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2029ebf4dd778b25d084af4e1383c8952caf0bd2f0c8e5940d2f82be7158e34d."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4b266bce7d640560951da1ff3eda5e0cc1550c8614b2a1ea6183519c8d722903.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 468c19abe84ce7bca229d73c81d4ded6564c8d148b95f351a057e04b1b80098e."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fb8eb364963656a48995d4052deca74cec8dfd07395d74788b843d7264a7128.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5696e1a6c73da853e281aa34b7d6d2fe6dc8be97c211b86c2896a53e6526045a."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75ac82bc1ef628d35241da2b78df5701adbe94efd1da970a2c40b47deb921515.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 193a37484a4f7f25aa08e90c23783e2bbacb3cac45721e59fc60bd86d6b12c79."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b9455f8cdc76ec4612f947ce914f722c74032563c11075dad9caf80353c14623.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b8cf58c826726dd7c44c2c15f2530af1e33cbe393bd65e8de867ef14ed2276c3."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a19272331578b9876a758d37c2e72dcac32cc7151aa0b2aeb549388f74c8cb51.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k128", "row", "col", "satfinite", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k64.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k64.json
new file mode 100644
index 0000000000..b3ec1f623a
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int4-m16n8k64.json
@@ -0,0 +1,146 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34df0473cf3a71f266d58df0459e952a3e93d223f7635e3c0d81fdec210e8e0a."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 884148218b4dbe52217fe5aaceae1dbe2a6322f13e2861aba575e5041cdcb472.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4_u4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e12664cd3c2c8ab1ebee94128e90959dfd9fde36f4023cd59ad4ae6b317a1107."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14dcc3ca9825b8567bc0ab8f60df13f8a003993172ec9d8b6463f31eb12b5b1e.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1a94dd1d6973c3bba56fbbb5e2972c81d74f9afa874e0e6435de228453839aa4."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b66dbbbbdb29f1a7d3e5cf07a95446ba1875fee261bb0b3e37206aa5c95c9e4d.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4_s4",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 206e31e80c020079ad30648d80a62afaf3564e99b9fc60bce10a6d013fe6c3a2."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 22fa426e26b6e51a27796d155fc1fe63e70a16bf4b36d73020d02c786769cd5e.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ef0b5eecdcc862411ed22bc193ddbcea8d94eb7d7f89158a7764a9759137fbc."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 782d840202c153255ac2ac63fdee9b5284b1fb992ec0e82e4fb388e11ba09780.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 940e92ff27f876255377efa918c901fed7e3e68e5ea77b987c1e267dc681f894."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4b51c1e1828a81d1cf6c7fe583ca851ab460871807158be6476b77382dedf9e.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06528543fb56e5d6e216226520c60e3fba9c10caee88988d660dc3bf62cfc402."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d5cab5865f1616c640a859d6acfd3d704c8a4b6a0504edd360b513973809e3.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u4", "u4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4.s4",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f81cee09a2ed0f8c2a26a84a608b4d8363fb89c6da62fbc53ef36e4669d56432."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "artifact_kind": "cubin", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb9ce5c9f7687237a762d8ad90791c121859cc83620ca6ef2755a0ead7e93138.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u4", "s4", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8-m16n8k64.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8-m16n8k64.json
new file mode 100644
index 0000000000..6867181651
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8-m16n8k64.json
@@ -0,0 +1,146 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 978ee747ff8cc5c08d711a0c53853888946906e2669cc24a21928ad8adf71e8e."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 018e9891184767f6e465cb1f24794c6c0145ef5077c2beeca95e42e05aa4189d.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8_u8",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b74c85eca5faf76b32dcda134941e5b09e74cf1aadbfca395a09ad0a4f66286a."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 041f6695d70e0ae239ca5a008247c1c84c651ecdff45cb9a578559f6c0a27731.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4aa071df474c2e447be1941f74662bbb1420c01e4fdab1c071aca9d571863b6f."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ca07c7e22bb41deec0862f637e40df38b910149068f4abc8a1e480493439e7f3.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8_s8",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 72e7739e891781c3475cf8cacdf67a739e6623660d0f83794818612cd297a82e."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a7361b01c024ab1c79cbe8744f497eaeba5eacbe7ead475ba79e21b317abe01f.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2ae77445ff0a92724257c8a3fb55f6db72266289f6b97e6b47d21516466e5821."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e6527d661a654f0d5d7b508d529f9008ea0aa02aca95bc47712d9669b74a4988.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c147cb0046175cd363c824b78ace3b1acdd06bb5c9d49990db97a715a2ef530."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3859169c6918f08e624d5ace5c0745797f3757d9f96cad14b37655e71e37e405.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b3d041020ef9ee6fb6013cc491495ec4add4a035da239b9d560b1017e0daeda7."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2b1813cc3f516d79a19fa139c7fc5d119fc1392db6b1f299fb6ef0ef46df6ca.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    },
+    {
+      "id": "mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite",
+      "source_record": "int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8_s8",
+      "llvm_symbol": "llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8.s8",
+      "llvm_arguments": ["i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32", "i32"],
+      "llvm_results": ["i32", "i32", "i32", "i32"],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx85",
+      "status": "validated",
+      "stages": [
+        {"targets": ["sm_80", "ptx85"], "representation": "separate probes for the recorded lowering mechanism", "stage": "backend_codegen", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 309f99b05e8c94a7d67043662d247b6d2181ace2fa180f341d891848fe49df87."},
+        {"targets": ["sm_80", "ptx85"], "representation": "separate PTX outputs assembled for the recorded target", "stage": "ptx_assembly", "mechanism": "inline_ptx", "outcome": "succeeded", "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1677a6dc99df6612f9de3109d52d45b0293847849900d5ed579f651df96031ea.", "tool_path": "/usr/local/cuda-13.3/bin/ptxas", "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0", "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"}
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k64", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8.json
similarity index 53%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8.json
index f8c707a173..5be7696059 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-sparse-mma-ordered-int8.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-ordered-int8.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "mma_sp_ordered_metadata_m16n8k32_s32_s8",
@@ -18,21 +18,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 signed INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6fd14bda0a191ba64fe0a74811fd81acef22dae9a7364c814feb5dc6b82f25b6."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33a756465fa9b33d89f2b1ab0da1d5ce7a1f0e8293d97888b8db3e9cdc7a1f4a.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -49,21 +49,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 64edac3534d432072fb0cb9ee94ae758f539d41f98556515b3c8269967feb7a6."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9111691371ac05725c5828e7342d57b2c0dc17a189490d5fe06a4e499c657703.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -80,21 +80,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 unsigned INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cd7e55a90134cfc35729cd87003a44c454b151e8910822841d1d18f0c9ebc1a2."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1566d75ad4047bc212b72c1d4e10c883b8616bb4276963654b1d5a2086c2aa8.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -111,21 +111,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12003882bbbb7dab7f4bec48ce3a009d0b9d08e185f69fb277afceef476fb7c9."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e0029151f55c5f3974fcf5c52523d22289796d7bfbd8fa74916f3cee0a82459d.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -142,21 +142,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 signed saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0097b025872046cef28efa3c9f463bf8993bb9c41da22afc8477efc2a17a4ac3."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75f59d33a1fc63c903fefc73ee06a31dfa1364d3a5dedcaa5cda933152cc92d2.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "s8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -173,21 +173,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b93a06888a9596b56f9ec7dec0e30eb7a87767aa6c4bab12aac88b0c80e4c7e0."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6fb8d4d5c0e41f500f6c523bd132db310bb25628a68ae9cb1331f0a890017d87.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "s8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -204,21 +204,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 unsigned saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c30b514d66dd4ef7d3898ed560a1144be98a172ef526ce4388beaf1f7c7dbd44."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 766d57b5690df60d31209316dc540cfc2f8c31c303b3b3fae6d523385f764f40.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "u8", "u8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
@@ -235,21 +235,21 @@
       "status": "validated",
       "stages": [{
         "targets": ["sm_80", "ptx85"],
-        "representation": "convergent ordered-metadata sparse m16n8k32 mixed-sign saturating INT8 inline PTX with selector immediates 0 and 1",
+        "representation": "separate probes for the recorded lowering mechanism",
         "stage": "backend_codegen",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80."
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c10be1e6711b2a58cbe3fa9cf60de15556b9496ba056edee6036a3c20b8dd2d6."
       }, {
         "targets": ["sm_80", "ptx85"],
-        "representation": "exact-floor PTX for ordered-metadata sparse m16n8k32 INT8 MMA",
+        "representation": "separate PTX outputs assembled for the recorded target",
         "stage": "ptx_assembly",
         "mechanism": "inline_ptx",
         "outcome": "succeeded",
-        "detail": "CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80.",
+        "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c07f0d6c2eea06bdaa939cf307a9323910485971dd341374537c4067524534c.",
         "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-        "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-        "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+        "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+        "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
       }],
       "runtime_validation": "unexecuted",
       "expected_ptx": {"mnemonic": "mma", "modifiers": ["sp::ordered_metadata", "sync", "aligned", "m16n8k32", "row", "col", "satfinite", "s32", "u8", "s8", "s32"], "operands": [{"kind": "register_list", "length": 4}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 2}, {"kind": "register_list", "length": 4}, {"kind": "register"}, {"kind": "immediate"}]}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-standard-int4.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-standard-int4.json
new file mode 100644
index 0000000000..82abe1c23a
--- /dev/null
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-sparse-mma-standard-int4.json
@@ -0,0 +1,1570 @@
+{
+  "schema": 5,
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
+  "backend_kind": "llvm_nvptx",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
+  "records": [
+    {
+      "id": "mma_sp_m16n8k64_s32_s4",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ae995db003010942ebdf8a541fb71e7a209b14d2035dc9b39ceebb44a20a8b22."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d3e0072eed343ab18506217d990ecc273e425433b2c7866bbf1cb84c84c48651.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "s32",
+          "s4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_s4_u4",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.s4.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 512623d519f30b4de0fca57417959c5a3e4f6208ac12e3691744f3fbec21cefa."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e549ea726768ab22519dacc89c24c3546e6cfcf709f8b53b5a8e86030aa8deb.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "s32",
+          "s4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u4",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 48f8afefe6dcc736f9b1e6750a9aa96bcbe0f1406d48a768f5ec04527b9f8ff2."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e41e8f8421b3066da06e8b9f77e52d19c248e2833be81e57cbe83b1dc4e3b34a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "s32",
+          "u4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u4_s4",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.u4.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cfc39a65a680d1cbbc4c17da9f23149ff569b977c427d1b56d9e22ae71752d6d."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c523935c6e4d65c559d3b3246acf09a9515e0fade43ca29a79225e3009374.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "s32",
+          "u4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0d1ce7e2d45fa78ac932971fa7ccc0f46e00d454de87b3e1ea926b0b22636b9b."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f8e6cb0ffef1b7c51d3caa4e5f589992e341d54ef3c6b0f0d1e8ac2581ae5db2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "s4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_s4_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 402335a93decd5671794a0c5ec5e6f09ffd69ed6b26754fac45e7f294b3aac25."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11e848685fa3b253ee0b3abb5fee26cf6fb7b3e7f567351483da94bd54ed1afc.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "s4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d2ab75eca2756a0c4d2c5544fe7837b92630aa42748c757bf68e0dcf6cc0c8f."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dea3c30a40b3bf4e28518ac19ce412aed32cb5f0394d48bb75ee09953527c0f4.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "u4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k64_s32_u4_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c8888ee61f060e895057d138bf3de9ec113eb39bb2c874055eda73abac1316e9."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 770bae1810b48f7e160fb5a12fc8c49e4075d0c9cf30402bfa707c38346aa36a.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k64",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "u4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 2
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_s4",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f178dc6498dfe1335f303815c77a74f6c64eef9102c2ae74312466d8ac126ab0."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e32824932666209d5959192f45f412c7f087fd40cd3b8a97e82e9004721855c5.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "s32",
+          "s4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_s4_u4",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.s4.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 751f4ccde1a953061a6de9f03fc03e29bcab215d3e0dfaf5cadec2fdab2ffda9."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4b7f6a5ffd37de7a86d2c35f122c69356b19e3ab05c2eaf64e726dd677fffd2.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "s32",
+          "s4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_u4",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa743b0396151d7184adfa37f3ff722cd693b7bdb74f433a522498874947fbfb."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 56253d8e8cd56f03dadc8369add61af230eac577a3258c42a28dcf4f2ede4f3d.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "s32",
+          "u4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_u4_s4",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.u4.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6c8d0ebc3993fe3649feb8dc574ca4aa6d9bbb8a54b57bab64439c72b3183f91."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e93fae493c1bff6c15d4b462fe8d3e9926f99da8c339f07247d5633515195e36.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "s32",
+          "u4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f05e41f54c80df9399f1535e4630d9ac644c42c0c8bb66c19902effd9fe96f3b."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2733247c94953620e2bdfdda7e22bb9f6a0f731285caa41a50d70a50d63c92f3.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "s4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_s4_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e91aa50a469e4ae22cabf614f0c86c11a381dd99b7d57a4f376c7f86b815819."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c0e9109eafe381fc83456148a9518994753df9868388321037f0cbe11c64a260.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "s4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_u4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f13fef29cc4b67af7e5a181bef0b806732d7eea0ec0f5176e91581b982f20511."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0557fe188cb6b3e8cf492411a8e5ae3072d44fd5ab349e21951c62b0ed30a9.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "u4",
+          "u4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    },
+    {
+      "id": "mma_sp_m16n8k128_s32_u4_s4_satfinite",
+      "source_record": "int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4_s4",
+      "llvm_symbol": "llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4.s4",
+      "llvm_arguments": [
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "llvm_results": [
+        "i32",
+        "i32",
+        "i32",
+        "i32"
+      ],
+      "target_triple": "nvptx64-nvidia-cuda",
+      "gpu_target": "sm_80",
+      "ptx_feature": "+ptx71",
+      "status": "validated",
+      "stages": [
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate probes for the recorded lowering mechanism",
+          "stage": "backend_codegen",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75d63c6f58dbf64b7ff7c0bff1b89fd0c2c7776ecc86997ccfcd3184130f610f."
+        },
+        {
+          "targets": [
+            "sm_80",
+            "ptx71"
+          ],
+          "representation": "separate PTX outputs assembled for the recorded target",
+          "stage": "ptx_assembly",
+          "mechanism": "inline_ptx",
+          "outcome": "succeeded",
+          "artifact_kind": "cubin",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3f4ba91d089260787eb84c1e93a4b9ad6ea30660ea314e755f025574e205657.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
+        }
+      ],
+      "runtime_validation": "unexecuted",
+      "expected_ptx": {
+        "mnemonic": "mma",
+        "modifiers": [
+          "sp",
+          "sync",
+          "aligned",
+          "m16n8k128",
+          "row",
+          "col",
+          "satfinite",
+          "s32",
+          "u4",
+          "s4",
+          "s32"
+        ],
+        "operands": [
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register_list",
+            "length": 4
+          },
+          {
+            "kind": "register"
+          },
+          {
+            "kind": "immediate"
+          }
+        ]
+      }
+    }
+  ]
+}
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-special-registers.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-special-registers.json
similarity index 52%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-special-registers.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-special-registers.json
index cb97ba146b..52b340537a 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-special-registers.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-special-registers.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [],
     "concrete_llvm_arguments": [],
@@ -19,23 +19,23 @@
       "stages": [
         {
           "targets": ["sm_20", "ptx32"],
-          "representation": "side-effecting inline PTX reads of smid and warpid",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted both volatile location samples in 633-byte PTX with SHA-256 8b6bdea0bf90c68baecd7189ae84ba5c24a0ca32a3a347a1dc72ca8a0a62277f."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ec007869a74a8d4d9bf7be114f276c2c4e26f3e4c34eeaab721362bbf9a1918."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX retaining smid and warpid reads",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 2,984-byte cubin with SHA-256 2a1f9febb72d01384c5c1c4583e4511020b2c086144d8b25f6aad354a4c34618.",
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14f47ee4cc2165c26a56b268fb647e325de01ae94e284257de734b97dea0626a. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -45,23 +45,23 @@
       "stages": [
         {
           "targets": ["sm_20", "ptx41"],
-          "representation": "pure inline PTX reads of dynamic and total shared-memory sizes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted both size reads in 655-byte PTX with SHA-256 4ab2630ee33912457ed45c1cd516ae0797e062fcba7123db9bac352c67280b3a."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24cc1707db68cd33a711956ac50c8c65ab3e84eb5cf71a00fc15160d770b645d."
         },
         {
           "targets": ["sm_75", "ptx41"],
-          "representation": "LLVM-produced PTX retaining both shared-memory size reads",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 ee3358b59d2b596f1b276d983c3149d58c384872ee5edad482cf20a99183e7b1.",
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7feb7ce1442470cb7ca3f613338c9fa901e16b23bc621a84ac3aa46a8583088d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx41, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -71,23 +71,23 @@
       "stages": [
         {
           "targets": ["sm_30", "ptx32"],
-          "representation": "pure 64-bit inline PTX gridid read",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted the full-width gridid read in 582-byte PTX with SHA-256 ee2f3fba8b51cd03bace8fccd9e3a76817240b7eb6e69e29943df3838b730159."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fb0578258494bd20b8dc20c6fe2243954b382980ac39ab5577a752f9af9a20e5."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX retaining the 64-bit gridid read",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 39c56630ff2afdadeb521c2a56023f289a2a39f0d39b0d01e2a5b3d9171dedd9.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0236c2ff2b8a9c74dc5ef689d53ae080fbc93f5d23cee0b18105663b14080f9c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -97,31 +97,31 @@
       "stages": [
         {
           "targets": ["sm_20"],
-          "representation": "typed declarations for clock, clock64, envreg1, envreg2, nsmid, and nwarpid",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0."
+          "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385."
         },
         {
           "targets": ["sm_20", "ptx32"],
-          "representation": "typed calls lowering to six exact special-register reads",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65."
+          "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX retaining all six typed special-register reads",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25.",
+          "detail": "Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -131,31 +131,31 @@
       "stages": [
         {
           "targets": ["sm_30"],
-          "representation": "typed globaltimer declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis produced 571-byte canonical IR with SHA-256 05f5c3e1cb6db2e75012139ce279738ff495d4073aacfd429b50bdfde6153f44."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 397d8179a2161540732e23d2358db3f5b9d44664430b730aac91c4dc83e5cded."
         },
         {
           "targets": ["sm_30", "ptx32"],
-          "representation": "typed globaltimer call lowering to an exact 64-bit read",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Pinned llc emitted the read in 548-byte PTX with SHA-256 a78d9cae78923fa3a3be7909f47e29aedb6bfa663681173ec7a9aed44e8f4bfe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7610b182b5f89de1c960aeb1cac1a4aff14a652935f535dc6e90da6fb8a211e8."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX retaining the typed globaltimer read",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the virtual sm_30 PTX and produced a 2,856-byte cubin with SHA-256 70b45763f48528285ad1525d565230adce99b4abc193a0f6302ca82be86024b2.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5001111b2c815febac9426aed7a4b9a51ab752be9b2d97985f60a745e49cd0a7. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-stmatrix.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-stmatrix.json
similarity index 69%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-stmatrix.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-stmatrix.json
index 063ccd24c3..2d052958b6 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-stmatrix.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-stmatrix.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "concrete_llvm_results": [],
@@ -22,31 +22,31 @@
       "stages": [
         {
           "targets": ["sm_90"],
-          "representation": "four opaque-pointer typed llvm.nvvm.stmatrix declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "four typed LLVM probes retaining the x2, x2.trans, x4, and x4.trans stores",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1."
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764."
         },
         {
           "targets": ["sm_90", "ptx78"],
-          "representation": "LLVM-produced PTX 7.8 modules containing the four stmatrix forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843.",
+          "detail": "Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-control.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-control.json
similarity index 59%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-control.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-control.json
index 98fbe41123..a0e7ca794b 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-control.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-control.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "target_triple": "nvptx64-nvidia-cuda",
@@ -24,14 +24,14 @@
             "sm_100a",
             "ptx86"
           ],
-          "representation": "three convergent inline-PTX kernels covering the admitted tcgen05 commit and shift controls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above.",
-          "tool_path": "llc",
-          "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4.",
+          "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+          "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
         },
         {
           "targets": [
@@ -40,23 +40,23 @@
           ],
           "representation": "isolated direct-intrinsic probes at and outside the imported selection predicates",
           "stage": "backend_codegen",
-          "outcome": "failed",
-          "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors."
         },
         {
           "targets": [
             "sm_100a",
             "ptx86"
           ],
-          "representation": "LLVM-produced PTX containing the three exact tcgen05 controls",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN.",
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
@@ -66,8 +66,8 @@
           "representation": "the same three inline instructions emitted for excluded family and future targets",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-cp.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-cp.json
similarity index 94%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-cp.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-cp.json
index c23cb6ca38..456531a241 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-cp.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-cp.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "target_triple": "nvptx64-nvidia-cuda",
@@ -24,26 +24,26 @@
           "sm_100a",
           "ptx86"
           ],
-          "representation": "two convergent LLVM kernels covering the 17 copy members separately for CTA groups 1 and 2",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a."
+          "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449."
         },
         {
           "targets": [
           "sm_100a",
           "ptx86"
           ],
-          "representation": "LLVM-produced PTX retaining all 34 copy forms in separate CTA-group kernels",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets.",
+          "detail": "Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-ld.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ld.json
similarity index 92%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-ld.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ld.json
index f1d2e77a00..2a900230f5 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-ld.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ld.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [
       "tmem_ptr",
@@ -27,11 +27,11 @@
             "sm_100a",
             "ptx86"
           ],
-          "representation": "58 convergent inline-PTX kernels covering every raw and pack16 tcgen05 load leaf",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures."
+          "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58."
         },
         {
           "targets": [
@@ -41,23 +41,23 @@
           "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6."
         },
         {
           "targets": [
             "sm_100a",
             "ptx86"
           ],
-          "representation": "LLVM-produced PTX retaining all 58 raw and pack16 load kernels",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads.",
+          "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
@@ -67,8 +67,8 @@
           "representation": "LLVM-produced PTX retaining the same 58 load instructions on an excluded architecture",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-ldst-offset.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ldst-offset.json
similarity index 91%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-ldst-offset.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ldst-offset.json
index fdd25efc3c..ca7744240d 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-ldst-offset.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-ldst-offset.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "target_triple": "nvptx64-nvidia-cuda",
     "gpu_target": "sm_100a",
@@ -23,14 +23,14 @@
             "sm_100a",
             "ptx86"
           ],
-          "representation": "32 convergent inline-PTX kernels covering every raw/pack16 load and raw/unpack16 store leaf for tcgen05 16x32bx2",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16.",
-          "tool_path": "llc",
-          "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+          "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6.",
+          "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+          "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
         },
         {
           "targets": [
@@ -40,23 +40,23 @@
           "representation": "the same paired fixture requested below the effective target floor",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0)."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0)."
         },
         {
           "targets": [
             "sm_100a",
             "ptx86"
           ],
-          "representation": "LLVM-produced PTX retaining all 32 paired offset kernels",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10.",
+          "detail": "Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
@@ -66,8 +66,8 @@
           "representation": "the same paired PTX fixture with target spelling changed to excluded plain and future targets",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-mma.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-mma.json
similarity index 61%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-mma.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-mma.json
index 1606d92959..0f199bdec0 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-mma.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-mma.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-tcgen05-mma",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-tcgen05-mma",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "target_triple": "nvptx64-nvidia-cuda",
@@ -20,174 +20,174 @@
       "stages": [
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed.",
-          "tool_path": "llc",
-          "tool_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-          "tool_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921"
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad.",
+          "tool_path": "/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc",
+          "tool_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+          "tool_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449"
         },
         {
           "targets": ["sm_100f", "ptx88"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69."
         },
         {
           "targets": ["sm_101a", "ptx86"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea."
         },
         {
           "targets": ["sm_101f", "ptx88"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c."
         },
         {
           "targets": ["sm_103a", "ptx88"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323."
         },
         {
           "targets": ["sm_103f", "ptx88"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51."
         },
         {
           "targets": ["sm_110a", "ptx90"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6."
         },
         {
           "targets": ["sm_110f", "ptx90"],
-          "representation": "24 generated inline-PTX probes plus direct kind-selector oracle probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1."
         },
         {
           "targets": ["sm_100a", "ptx85"],
           "representation": "direct intrinsic probes below each paired target floor",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe."
         },
         {
           "targets": ["sm_101a", "ptx90"],
           "representation": "direct intrinsic non-monotonic PTX negative",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0."
         },
         {
           "targets": ["sm_120a", "ptx90"],
           "representation": "direct intrinsic excluded-target negative",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_100f", "ptx88"],
-          "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_103a", "ptx88"],
-          "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_103f", "ptx88"],
-          "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_110a", "ptx90"],
-          "representation": "CUDA-assembled PTX from all 24 generated probes and four direct selector probes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_110f", "ptx90"],
-          "representation": "CUDA-assembled PTX from all 24 generated probes and three direct selector probes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": ["sm_101a", "ptx86"],
           "representation": "CUDA assembler target-spelling check for LLVM-emitted sm_101 PTX",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-st.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-st.json
similarity index 92%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-st.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-st.json
index 3cb27a875f..c328f12a25 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05-st.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05-st.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "target_triple": "nvptx64-nvidia-cuda",
@@ -24,11 +24,11 @@
             "sm_100a",
             "ptx86"
           ],
-          "representation": "58 convergent inline-PTX kernels covering every raw and unpack16 tcgen05 store leaf",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures."
+          "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373."
         },
         {
           "targets": [
@@ -38,23 +38,23 @@
           "representation": "the same fixture requested below the reviewed tcgen05 PTX floor",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6."
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6."
         },
         {
           "targets": [
             "sm_100a",
             "ptx86"
           ],
-          "representation": "LLVM-produced PTX retaining all 58 raw and unpack16 store kernels",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores.",
+          "detail": "Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
@@ -64,8 +64,8 @@
           "representation": "LLVM-produced PTX retaining the same 58 store instructions on an excluded architecture",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
-          "outcome": "failed",
-          "detail": "CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
+          "outcome": "not_revalidated",
+          "detail": "Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
           "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
           "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05.json
similarity index 92%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05.json
index c846cdf085..96abb1aa69 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tcgen05.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tcgen05.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "target_triple": "nvptx64-nvidia-cuda",
@@ -24,26 +24,26 @@
           "sm_100a",
           "ptx86"
           ],
-          "representation": "one convergent LLVM module containing the exact inline-PTX carrier for all 24 generated tcgen05 routes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16."
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6."
         },
         {
           "targets": [
           "sm_100a",
           "ptx86"
           ],
-          "representation": "LLVM-produced PTX containing all 24 exact tcgen05 forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract.",
+          "detail": "Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-threadfence.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-threadfence.json
similarity index 53%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-threadfence.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-threadfence.json
index bf89219ab0..36db2778a7 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-threadfence.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-threadfence.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_arguments": [],
     "llvm_results": [],
@@ -24,31 +24,31 @@
       "stages": [
         {
           "targets": ["sm_20"],
-          "representation": "three typed llvm.nvvm.membar declarations with no arguments or results",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209."
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a."
         },
         {
           "targets": ["sm_20", "ptx32"],
-          "representation": "three typed LLVM probes retaining membar.cta, membar.gl, and membar.sys",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3."
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a."
         },
         {
           "targets": ["sm_75", "ptx32"],
-          "representation": "LLVM-produced PTX 3.2 modules containing the three membar scopes",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma-descriptor-control.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-descriptor-control.json
similarity index 75%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma-descriptor-control.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-descriptor-control.json
index 7dd6a265e6..9d6424ad58 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma-descriptor-control.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-descriptor-control.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-tma",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-tma",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "concrete_llvm_results": [],
@@ -21,31 +21,31 @@
       "stages": [
         {
           "targets": ["sm_90"],
-          "representation": "11 individually generated typed TMA prefetch declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR."
+          "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714."
         },
         {
           "targets": ["sm_90", "ptx80"],
-          "representation": "11 typed TMA descriptor and tile-prefetch probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe."
+          "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466."
         },
         {
           "targets": ["sm_90", "ptx80"],
-          "representation": "11 LLVM-produced TMA prefetch PTX modules",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+          "detail": "Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -55,31 +55,31 @@
       "stages": [
         {
           "targets": ["sm_100a"],
-          "representation": "two individually generated typed gather4 TMA prefetch declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized both declarations. SHA-256 over sorted per-operation ID/hash lists is 4c60ef82171ca5e9bdd92a90683367bbb73023593d27e1af329fa9af9cbce511 for source IR, d186a1e64a5832f4bbe70323a18796cfaef6ab5d4b685b32709085ff1c24e960 for bitcode, and 609578ee9ee7d3bc1bb280cdf57aaa6b7bb1ea1a341a159ff3dec47c0d235ec0 for canonical IR."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcf9f4e3b994a7ccfade3699fa2c7393855b07aee83a70d3c449d4f4260098c4."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "two typed gather4 TMA prefetch probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM emitted both exact instructions at their catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 d5149bf319d478db6e1e40978c5acb62203fbb92fa5d767f0a4ddeaa3eb8fb8e."
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 701c7338ba48b42868907f3900d1074b019fab8e257f5676f9a92e368c4718f1."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "two LLVM-produced gather4 TMA prefetch PTX modules",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted both modules; the sorted per-operation ID/cubin-hash list has SHA-256 3a914c3e352965013e8a3b1a937f4faa4b4966830f04ca63a3c4d982427cf3d5.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+          "detail": "Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0635d30976c9afa0d905c53f55599ce0fa36a009828f00a5ae0f5283857a21de.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -89,31 +89,31 @@
       "stages": [
         {
           "targets": ["sm_90"],
-          "representation": "eight individually generated typed tensor-map fence declarations",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR."
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b."
         },
         {
           "targets": ["sm_90", "ptx83"],
-          "representation": "eight typed tensor-map proxy-fence probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79."
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b."
         },
         {
           "targets": ["sm_90", "ptx83"],
-          "representation": "eight LLVM-produced tensor-map proxy-fence PTX modules",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+          "detail": "Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -123,23 +123,23 @@
       "stages": [
         {
           "targets": ["sm_90a", "ptx83"],
-          "representation": "ten exact inline-PTX tensor-map replacement probes",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552."
+          "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc."
         },
         {
           "targets": ["sm_90a", "ptx83"],
-          "representation": "ten LLVM-produced tensor-map replacement PTX modules",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+          "detail": "Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -149,23 +149,23 @@
       "stages": [
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "exact inline-PTX tensor-map swizzle-atomicity replacement probe",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM retained the exact instruction in PTX; the ID/PTX-hash list has SHA-256 588486907e9e70101b6445f23f07d27798a009766a831e519c21a0a2fe4ea178."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 02683e3a77ca0ae4f160645826b8e3da2b9da126376f94ce15265ad34e8ca282."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "LLVM-produced tensor-map swizzle-atomicity replacement PTX",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module; the ID/cubin-hash list has SHA-256 4bfa0ad94eac3615d96e64ef7594b7b7b96b549057da9dd83f38afd135c5aed1.",
-          "tool_path": "ptxas",
-          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73",
-          "tool_sha256": "ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f"
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9d36f1191de4ceed2adb6d065c793c92620591a4fbd3bafe95571a43268f4ec3.",
+          "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma-reduce.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-reduce.json
similarity index 98%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma-reduce.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-reduce.json
index eac7354f3d..e9532302eb 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma-reduce.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma-reduce.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-tma-reduce",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-tma-reduce",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "concrete_llvm_results": [],
@@ -24,36 +24,36 @@
           "targets": [
             "sm_90"
           ],
-          "representation": "64 per-operation LLVM modules covering every cp.reduce.async.bulk.tensor typed NVVM intrinsic",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode."
+          "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66."
         },
         {
           "targets": [
             "sm_90",
             "ptx80"
           ],
-          "representation": "64 typed LLVM calls covering the complete TMA tensor-reduction matrix",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877."
+          "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc."
         },
         {
           "targets": [
             "sm_90",
             "ptx80"
           ],
-          "representation": "64 LLVM-produced PTX modules assembled for sm_90",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920.",
+          "detail": "Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
           "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma.json
similarity index 79%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma.json
index ab4c744e30..1a8e556197 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-tma.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-tma.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-tma",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-tma",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "concrete_llvm_results": [],
@@ -22,31 +22,31 @@
       "stages": [
         {
           "targets": ["sm_90"],
-          "representation": "one LLVM module containing the 14 Hopper TMA declarations and calls",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83."
+          "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581."
         },
         {
           "targets": ["sm_90", "ptx80"],
-          "representation": "typed LLVM calls for every non-CTA-pair TMA operation",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a."
+          "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a."
         },
         {
           "targets": ["sm_90", "ptx80"],
-          "representation": "LLVM-produced PTX retaining all 14 Hopper TMA instructions",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c.",
+          "detail": "Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -56,31 +56,31 @@
       "stages": [
         {
           "targets": ["sm_100a"],
-          "representation": "one LLVM module containing the CTA-pair TMA declaration and call",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as canonicalized the intrinsic attributes. The 521-byte source had SHA-256 c8f988c762ae488faa6a7a31f1391a74b33918a088449f4ec2bfbdb1ca33a201; the 1,964-byte bitcode had SHA-256 dd23f8a1a678731e95b7e2283da19530bce758b7b55ded0c98d01e99c971b978."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 065ffbad7cc3f4a868f3de9e7f86d3a4fb1b0ea7047e1ddcdd578192e1fe101c."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "typed LLVM CTA-pair TMA call",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned LLVM emitted the exact cta_group::2 instruction with no extern call. The 1,098-byte PTX hashes were 88d30f75f80d7603ec5852572942e3406db97d10a49bdb7b65c64fa1bcea2257 for sm_100a, ed24ee1e914179959d69b7b153d5d65bbead0c864bfeb624871f9c19f5cc17f3 for sm_101a, 451331c3b63021119d0e9ccdc7617e70b505943c69e1066d193ef1b74b54523e for sm_103a, and 13c7be98f933799fafa4055423395bc1d5d4c2f4303e70f27c38ded9a0e1ace8 for sm_110a."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6f52db1ccb491246b76348fb43a73fc961823d087d80432a2cec08dc36439f1."
         },
         {
           "targets": ["sm_100a", "ptx86"],
-          "representation": "LLVM-produced PTX retaining the CTA-pair TMA instruction",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas produced cubins for sm_100a, sm_103a, and sm_110a with SHA-256 7c522a3420f9856b526e23c6e3dc4067d244b27095057b91ed1b4af27a8abbcf, b3e3934adc407c847b428d93936f7d8e170059f4004218774c8755f71d599419, and ef654bdc632b1887f8622e765f8f5a2d0571b97c28eb39f17fd355483d8b9d31. Separate probes rejected sm_120a and sm_121a because cta_group::2 is unsupported.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eea1fea3e3abbce7a6d377fafd8ad8e2c8fec998a990afbd664b7ce4f3856dff.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-wgmma-control.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-wgmma-control.json
similarity index 58%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-wgmma-control.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-wgmma-control.json
index 87fe302d4d..d3b30f1c8f 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc-wgmma-control.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0-wgmma-control.json
@@ -1,10 +1,10 @@
 {
   "schema": 6,
-  "backend_profile": "rust-llvm-23.1.0-16696adc-wgmma-control",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0-wgmma-control",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "defaults": {
     "llvm_results": [],
     "concrete_llvm_arguments": [],
@@ -23,31 +23,31 @@
       "stages": [
         {
           "targets": ["sm_90a"],
-          "representation": "typed declarations for all three WGMMA controls",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f."
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8."
         },
         {
           "targets": ["sm_90a", "ptx80"],
-          "representation": "one typed LLVM probe containing all three WGMMA control calls",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98."
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3."
         },
         {
           "targets": ["sm_90a", "ptx80"],
-          "representation": "typed LLVM PTX retaining all three WGMMA control instructions",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3.",
+          "detail": "Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     }
diff --git a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc.json b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0.json
similarity index 63%
rename from cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc.json
rename to cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0.json
index 50fc0ef80b..8cfa35d5ea 100644
--- a/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.0-16696adc.json
+++ b/cuda-oxide/intrinsics/evidence/rust-llvm-23.1.3-aab11ee0.json
@@ -1,10 +1,10 @@
 {
   "schema": 5,
-  "backend_profile": "rust-llvm-23.1.0-16696adc",
+  "backend_profile": "rust-llvm-23.1.3-aab11ee0",
   "backend_kind": "llvm_nvptx",
-  "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-  "backend_version": "LLVM version 23.1.0-rust-1.100.0-nightly",
-  "backend_sha256": "2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921",
+  "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+  "backend_version": "LLVM version 23.1.3-rust-1.101.0-nightly",
+  "backend_sha256": "7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449",
   "records": [
     {
       "id": "block_dim_x",
@@ -368,36 +368,36 @@
           "targets": [
             "sm_75"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3))",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d23e19589de35127bff359facfbc56c4c16a19fd1aeef8a4568c6080ad4c1a68."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16.p3(ptr addrspace(3))",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x1.shared.b16 PTX."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8eb444c30aad31f7ce8a70bcd06b91bae96ddd810e5c0e0bad67d669bba76463."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a38549cf45825e7cf005005dcaa96cbdff09dd83f2b672a3d5bc23b8102018e.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -449,36 +449,36 @@
           "targets": [
             "sm_75"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.trans.b16.p3(ptr addrspace(3))",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 078c3ec1a8767b2764c1a5dd2617806f672124e023a5c9fe5f3c40c6f5a36219."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.trans.b16.p3(ptr addrspace(3))",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x1.trans.shared.b16 PTX."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f448dfcfc8dc1e19e552c3f601db928004ab0f095c275cf846bc4e8adc3eceb."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a759f49ba62f6028179716e660082e3b36877cefbe37f9ad1aa91c27aebc22c.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -533,36 +533,36 @@
           "targets": [
             "sm_75"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3))",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2228a4bf5e0bbaf7685dcbd5f0e890507df94d30aa9165cbe879037b84c74a4e."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16.p3(ptr addrspace(3))",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x2.shared.b16 PTX."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c86c69e86d0084dd76afccfa7dd76416f5a906eda8f245fd5eaadfb94eeedfea."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b42051503a9dcfd9cab9b052388ae403c55c7b81a776697ba93d8146a265b42a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -616,36 +616,36 @@
           "targets": [
             "sm_75"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3))",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b6c14f6dd0fdb52ad2e22873b19609ff47a1c8994bc5e3ce39c88bdc59a58f."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16.p3(ptr addrspace(3))",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 PTX."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51c2def02e859676a80a71edbbd6563d7dac75a59fb25899f3718a20705b267d."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67149c3d0b2f90b47e4e1f3f361f390adea9ffd4186e04bafaabd93b8091168f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -704,36 +704,36 @@
           "targets": [
             "sm_75"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3))",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28e76c919b457758f8eeaf5f976e581db31431ae31ed1efbc30a050a31da1344."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16.p3(ptr addrspace(3))",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Emitted exact ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 PTX."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c74dd5efb5771ae005406a00b33baa93333dd3fbcaa2787d0dee90957c48871f."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5dbfbc0015f15b42f13d4bb3e15a049ffe769e096166c94bbb0aecfaa567d43.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -792,36 +792,36 @@
           "targets": [
             "sm_75"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3))",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The recognized bare declaration materializes convergent, nocallback, memory(argmem: read), readonly, and captures(none) attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 585528599fab5e664c529894e42a2b9604bc6e503f9eefc9d2117235a0fd35fd."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "opaque-pointer typed intrinsic llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16.p3(ptr addrspace(3))",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "Emitted ldmatrix.sync.aligned.m8n8.x4.shared.b16 with a four-register result struct."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3d0cc72dd11e3399dbfe72a912f451f36835e6c60ff54db70e1b6589b905dbbc."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM ldmatrix probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted x4 PTX for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c31b7c8cbef8fc5d164d732bfa611edc166ee875735791b5446e77f385f2f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -863,25 +863,25 @@
             "sm_70",
             "ptx62"
           ],
-          "representation": "side-effecting inline PTX atom.global.add.noftz.f16x2 with =r,l,r,~{memory}",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted the exact native packed f16x2 global atomic-add instruction."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7769e680b674aa15719338a21a264d9e6a8bd0e7f4a3f90192b4f5507f2472cd."
         },
         {
           "targets": [
             "sm_75",
             "ptx62"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM packed-atomic probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas (whose supported floor is sm_75) accepted the sm_70/PTX-6.2 backend output when assembled forward for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a37572911b7e3d97e3785be11b99068c4b530eac6b0a16d6d74d6a0c6a56702. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx62, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -922,25 +922,25 @@
             "sm_90",
             "ptx78"
           ],
-          "representation": "side-effecting inline PTX atom.global.add.noftz.bf16x2 with =r,l,r,~{memory}",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted the exact native packed bf16x2 global atomic-add instruction."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2537a05bf31f004570ad178a6290d201b70855af40a531e7bb3c4930907b72f."
         },
         {
           "targets": [
             "sm_90",
             "ptx78"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM packed-atomic probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact bf16x2 inline-PTX instruction for sm_90.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5d3e002fed008a798ac7c23848acfa14f680ffa01f2d4c11154a6b127f3b71ce.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -1302,36 +1302,36 @@
           "targets": [
             "sm_61"
           ],
-          "representation": "typed intrinsic llvm.nvvm.idp4a.s.s(i32, i32, i32)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and added its no-memory and speculatable attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 de5ba8b2a4efbc2ca94148badabcdf0d50913a97a1663fb623592433f484e88e."
         },
         {
           "targets": [
             "sm_61",
             "ptx50"
           ],
-          "representation": "typed call to llvm.nvvm.idp4a.s.s",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact dp4a.s32.s32 PTX at the native floor."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9c42148d5c5a275ca9ece93315e98eb0e9508a0b4e8cf0eba8cc761268f29c42."
         },
         {
           "targets": [
             "sm_75",
             "ptx50"
           ],
-          "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d9b7b74a68441f7f40d480ab430166c946704ea11caf1a7e9951b4726130bc65. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -1387,36 +1387,36 @@
           "targets": [
             "sm_61"
           ],
-          "representation": "typed intrinsic llvm.nvvm.idp4a.u.u(i32, i32, i32)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and added its no-memory and speculatable attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4565681a3a6f6f64ea89b63e96c44c1aef2d594b01f3c695eaa4d35147c250f."
         },
         {
           "targets": [
             "sm_61",
             "ptx50"
           ],
-          "representation": "typed call to llvm.nvvm.idp4a.u.u",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact dp4a.u32.u32 PTX at the native floor."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eec368e7d9686a4072001c8362bc69ea91a9b25fb950b50232244c96e72c0887."
         },
         {
           "targets": [
             "sm_75",
             "ptx50"
           ],
-          "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 159beb8496be094d32d2c6421e8993b91ee8e0d8b787fa5313af73ea5af66eac. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -1474,36 +1474,36 @@
           "targets": [
             "sm_61"
           ],
-          "representation": "typed intrinsic llvm.nvvm.idp2a.s.s(i32, i32, i1 immarg, i32)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration, immediate selector, and no-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cf14ce2498510d2ccdf8e26062d9dab33a2f018ab44d7402f731ee721edd9586."
         },
         {
           "targets": [
             "sm_61",
             "ptx50"
           ],
-          "representation": "typed call to llvm.nvvm.idp2a.s.s with selector false",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend mapped false to exact dp2a.lo.s32.s32 PTX at the native floor."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5004901d9a7fa9fc719218c051ffd4362bcef5a21c7bcca397e74fdb199f3733."
         },
         {
           "targets": [
             "sm_75",
             "ptx50"
           ],
-          "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c5f55a84bfd467a639f8fe3020d3fa05a6616160125d05ecd44cf631c78a954c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -1562,36 +1562,36 @@
           "targets": [
             "sm_61"
           ],
-          "representation": "typed intrinsic llvm.nvvm.idp2a.u.u(i32, i32, i1 immarg, i32)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration, immediate selector, and no-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6bba537191d91e4c7f577889e7832f879421a5a8b021a98982af2c068ecd5ac5."
         },
         {
           "targets": [
             "sm_61",
             "ptx50"
           ],
-          "representation": "typed call to llvm.nvvm.idp2a.u.u with selector false",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend mapped false to exact dp2a.lo.u32.u32 PTX at the native floor."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ad295ebd24c5fa94a1990f0e676049b7232bf0421f6419e2230f054510b87ede."
         },
         {
           "targets": [
             "sm_75",
             "ptx50"
           ],
-          "representation": "PTX 5.0 emitted by the combined typed dot-product probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aef5395fd4ebfc37a0a4fb468bb18c83a363528cd307bfeb7ed1218c5fd84fc2. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -1640,36 +1640,36 @@
           "targets": [
             "sm_20"
           ],
-          "representation": "typed intrinsic llvm.nvvm.barrier.cta.sync.aligned.all(i32)",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and added convergent and nocallback attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 25b383a09f3293d7f5ceef1bc66d8400702d7a5c68d73f24532c371030e69c11."
         },
         {
           "targets": [
             "sm_20",
             "ptx32"
           ],
-          "representation": "typed barrier call with fixed ID 0",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact bar.sync 0 PTX at its oldest supported target profile."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26b139fc6199abb20c9a7d92ddb0a30839fb2bdd3eafe285ec63bb49da4bb3c5."
         },
         {
           "targets": [
             "sm_75",
             "ptx32"
           ],
-          "representation": "PTX 3.2 containing bar.sync 0",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the emitted PTX when assembled for its supported sm_75 floor.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e287209ed91d6c476b48a3d08399a0911d64f601f3ae07e328db3f975fc2d2e9. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -1989,22 +1989,22 @@
             "sm_30",
             "ptx62"
           ],
-          "representation": "typed llvm.nvvm.activemask() declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, side-effect, and inaccessible-memory contract."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a17f68d96e936c632dbeb06ea74eae28b193c05cdfdc08373f0715f348eacf1."
         },
         {
           "targets": [
             "sm_30",
             "ptx62"
           ],
-          "representation": "typed zero-operand activemask call",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted activemask.b32 at the imported instruction floor."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 246910573ff59e4feb1a3b8880eb69d36c7a14f49debfe41f1207db7b4ea51d4."
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2185,36 +2185,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.bar.warp.sync(i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent and nocallback attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d5b275af3bb7cfa86185e4ce735308f68edfb3dd4a71261c8ed80131f8f67b84."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed masked warp barrier calls with register and immediate masks",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted exact register and immediate bar.warp.sync forms."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e7bf824b7fcf96ef357d4279091d73c4cd95081c9a10994d3a2abc4df06f58d."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing register and immediate bar.warp.sync forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted both forms and produced a 1,464-byte cubin (SHA-256 df48ead94aa2ad3402456419362350879e99993d4b2ce03a05f695e9d58fbf37).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 395cf308dc13aa17943c5d073be46dc163c23cd9add8444b68a0dcff491240de. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2264,36 +2264,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.idx.i32(i32, i32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c8015a293ea78a1394796bab968f8d4b065b433149894cf07283122ab75837d."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.idx.i32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.idx.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dcc151e0a9e33ce7a059a418221504ddf54bb43d6f1b8927cffd7cdd76c93939."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.idx.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0dab85a67df2645eacc4979c404b7be148188a35e648c7e180b82aba4269ae71. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2357,36 +2357,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cc205d40d08b4db407d73012b4e02bacca951cbf2407578220e1a9993241f7c4."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.bfly.i32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.bfly.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d1ab9e1df5d932129b785eb98a099581f503fadd7a7e22b4e7621f06dd4ecce."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.bfly.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f4465879f688e4239d1fc44f8e1c2008923c890cf59e0d1e2cad6653348622d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2450,36 +2450,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.down.i32(i32, i32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12f77b50bd9a2ac2e167e43862d4925890afbc944ce1be9abf7a42bef834a693."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.down.i32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.down.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e868aa2aec22540ec973c73eda5ad85c82e432f562a4956b89fc5ec681ff12c4."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.down.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 56902b3d564c19bb5e6ffca02b1f0f73c0c6ea1a440d7ba209fe9242e86d6b50. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2543,36 +2543,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.up.i32(i32, i32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d6fef706af3d2993713934e1f5156c47628a6c943899ad369edf140f7b9f5c57."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.up.i32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.up.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e9ac34403d0a172c008ea25d9e62a0786926818eec2efb2091811ddca6c0caf."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.up.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 faebc8c202740b059817f861bd3cded8188e653457a4b9791ce87506a1c63314. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2636,36 +2636,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.idx.f32(i32, f32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d782063e4b1ae122d58089206151ace8d5b7d4f660a5d5159fb1b770f56b46a0."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.idx.f32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.idx.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e4775aad2f0dbe6b4f3fc2db7d15985eed28a8adcc2031100d3db5a4c25600a4."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.idx.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 578376a2dccd0ed45a721f981ae2bf7094d65d90e46479f5569dc8a00354275f. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2729,36 +2729,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.bfly.f32(i32, f32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa450ce0f0bf68458c30deaa762636db4ffcfbd33d8f8c5d627a057b59740efb."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.bfly.f32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.bfly.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f67d1796c2477484afa60ecc1be1bc640a045537ba52e1512f1cb17bc46898e4."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.bfly.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4ab01d57f019c3f6994b977e6528bbadebb9db1980ff1e4e57480354c0cb1aa0. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2822,36 +2822,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.down.f32(i32, f32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fe998b216b523ddba326d894f7fd6c1c811b7b8f6793fd67fa9f32bb5426fe93."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.down.f32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.down.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e822e31752229f9d5f8379f58427407f4da67974fdf5cb800973a1d420d99843."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.down.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 78a64562571e8dfa2d02c779947163cd4c087bf0fb45cfc9b26b7a9448da1d34. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2915,36 +2915,36 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.up.f32(i32, f32, i32, i32) declaration",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a97a689d300dd7c199c714c01189afb154122960104cfc1a1f1e1180c4e7cf3."
         },
         {
           "targets": [
             "sm_30",
             "ptx60"
           ],
-          "representation": "typed llvm.nvvm.shfl.sync.up.f32 calls with register and immediate lane and member-mask operands",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted all four reachable shfl.sync.up.b32 operand forms with the fixed clamp."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a3ed93208637ba99cf7cdbdd4ddbcc7dd13f263fcecc0842c46753c8c841ca34."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 containing all four reachable shfl.sync.up.b32 operand forms",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted every generated register and immediate form.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 01942dd79db0ae31a8cb526a0bab0e1bbac0a00e9494f7fd2c1ac989af6a68b2. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -2992,25 +2992,25 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.idx.b32 instructions with shared operands, and reassembles i64",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 00eed2a3ce51403c06b4557ca323d691094cf59685823dd73b0b329a2fc40d9a."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM 64-bit idx shuffle probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1301d100cbe1985f544b748d01b5a92e32173c169be106074f0614f23f7a08d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3059,25 +3059,25 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.bfly.b32 instructions with shared operands, and reassembles i64",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a423fb03456e022c274ab0aee6f4480239603e12b7657f78a3cce23264b020e6."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM 64-bit butterfly shuffle probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41a4b0b1e404dde5ac1a6f457e13bde66b3a26ce037c64f3c691fa4666f07d5d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3126,25 +3126,25 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.down.b32 instructions with shared operands, and reassembles i64",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dc004e784b4ad1d4151d0fbbb8c7ab4815437672e35cb37c8408d685938d961."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM 64-bit down shuffle probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ac475b6fa0fb52beb3890ebaba00ef224041684fd47bfec82f609768fb0ad71f. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3193,25 +3193,25 @@
             "sm_30",
             "ptx60"
           ],
-          "representation": "one convergent side-effecting inline PTX block that splits i64, runs two shfl.sync.up.b32 instructions with shared operands, and reassembles i64",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a83c401208ac85d4c6fe20f942147f8bd8410442b4ce60b86079b138ceac2208."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the selected Rust LLVM 64-bit up shuffle probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a85d3dccb30f9491380b1f54913e7261c58667a5c22f195298c6a4f0738f3992. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3266,25 +3266,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX fma.rn.bf16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact fma.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f847702ab3d3db2051359e9ba8869735a57a5cf5b92a2901b154340dc14cc080."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa10765bb36d761080b94f2c0788a77e387b12032c06504740e5ff756239cda5.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3332,25 +3332,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX fma.rn.relu.bf16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact fma.rn.relu.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 903e4771d5a308e61a37282dd1f236481315e6519d5c06dbfd19c6c47f97acfd."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 adc9face773753523147665457351aa23a8b43da9e697a565109e15870383010.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3393,25 +3393,25 @@
             "sm_90",
             "ptx78"
           ],
-          "representation": "pure inline PTX add.rn.bf16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact add.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2c103ab626814fb233a6e15e90a48c2a6a5b78f7f13e38c4c8c6cebca28ef0d0."
         },
         {
           "targets": [
             "sm_90",
             "ptx78"
           ],
-          "representation": "PTX emitted by the combined three-form sm_90 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af1befea9f33a173ed1a34fa549dfd6e8cf85f450042ec4948cdcd9eef20051f.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3450,25 +3450,25 @@
             "sm_90",
             "ptx78"
           ],
-          "representation": "pure inline PTX sub.rn.bf16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact sub.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 379f54089c227f714a7a906db098eef11d0ac042db017cbd022236e2f366c440."
         },
         {
           "targets": [
             "sm_90",
             "ptx78"
           ],
-          "representation": "PTX emitted by the combined three-form sm_90 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e912f52acc5af50229ee7d87a37888ffbfd4dddc236e582d317512c281864d0.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3507,25 +3507,25 @@
             "sm_90",
             "ptx78"
           ],
-          "representation": "pure inline PTX mul.rn.bf16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact mul.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f4e083050e3dc987284bceea5b0064f86a0c78bca3754d8a92284911bbd7fe78."
         },
         {
           "targets": [
             "sm_90",
             "ptx78"
           ],
-          "representation": "PTX emitted by the combined three-form sm_90 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6f92a98210d4280598f8059312562b774c73bae8a2baf9af8252031d8f5ac5b.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3569,25 +3569,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX min.bf16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact min.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f59f875a1a686231cf0e0c09cd8ea1d7107aaf7ba547e54f0847ba4d71726b27."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51930c796a141b52e8382f91a54de007b9908945d7bdddf8085b3c9e4575557a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3630,25 +3630,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX max.bf16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact max.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1bcd79e15bcc5f861613af0657054804f2cc98de0bc44082c84eeb6e06750d11."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 37e990ab010dfd7a6aa363c1bc12abfb5d4e121bceb7b89ac8056454e95a2385.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3690,25 +3690,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX neg.bf16x2 with =r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact neg.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2dca2ee0088dbadc0cb0904d95659a2c0d5e7974da0c5e8aff75f0e290504a1."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 779e29da03300e6d8024f6c98694b6bbad23c127181f299430b3a6b09c8541b5.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3747,25 +3747,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX abs.bf16x2 with =r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted the exact abs.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fc97be133b128891fbdca9f6a4b9769cfd2712fa35ee81884e52b9b436f6ef32."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined seven-form sm_80 packed-bf16x2 probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbafd197892b5b499ef62752012ccb5e86318641dd9784942872a4ca7d685893.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3811,62 +3811,62 @@
           "targets": [
             "sm_80"
           ],
-          "representation": "typed intrinsic llvm.nvvm.ff2bf16x2.rn(float, float) -> <2 x bfloat>",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60)."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7b66943f3e72c942117544102ede74b7d46537dc9c8f09cb4c800b1a8150b25."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed call to llvm.nvvm.ff2bf16x2.rn(float high, float low) -> <2 x bfloat>",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 260159c4d2bde9577b33387b20e93e5e1e00605c7ce69ce157c5c5046bbaabf0."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c.",
           "artifact_kind": "cubin",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX cvt.rn.bf16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rn, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 625e5da377d73334b52ec4bdd6bc6702e15b196ca54350f0b070cb61e09b60ff."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -3913,25 +3913,25 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX fma.rn.f16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact fma.rn.f16x2 inline PTX in the combined four-form PTX 6.0/sm_70 probe. An independent typed call to llvm.nvvm.fma.rn.f16x2 emitted the same instruction at PTX 6.0/sm_70 and at its native PTX 4.2/sm_53 floor; sm_52 failed exact intrinsic selection."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e1cf777d5d76a7579b7c840ac1250cd7961888126aeef92b619e2bfcf5718aa."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f962661dab009881f23af0cdf33aed0bdb5ea13629f790897da04f271d8cc305. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -3979,25 +3979,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX fma.rn.relu.f16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67da91ea322e2d60af10838e8a20890d99c38e07262abe9a6848ea5e1fbbf2fb."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined three-form sm_80 packed-f16x2 upper probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 871b4ac65868041bcc0f028deaaac2793f5df9db4518fdbfaa03580e9c26414a.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4040,25 +4040,25 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX add.rn.f16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b7f49dda4585f12ddfc9ecaf71517120de1aefd78a9f843fcb511f81bbabaf4."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b33cc169935b856e4ff58245496b889d56c48c772653634f963facb395c24326. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4097,25 +4097,25 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX sub.rn.f16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 260eb7ac7c16c07d7eb3ac88e498b18d9e9f37b1c763d3dca78075888ea7cce8."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdc2a8724f570ba9cf569ea9240cf60a937116228c42a5d0d9bdcef1af591d35. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4154,25 +4154,25 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX mul.rn.f16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0642183f45157615161325bf3fd8c344f4d8efea483a8bf2c111fd9bd7c9dc4f."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX emitted by the combined four-form sm_70 packed-f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 090bddd2dbb7c5459d46de7c6956de9084a097a5041086b13a2a85322673072d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4216,25 +4216,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX min.f16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d54f22a9e49d8c8e53739fb3942a54749c50fe7749970cd67c4c77932fc1e11."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined three-form sm_80 packed-f16x2 upper probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34e9c8ebf06b8ee9b1df5eda184d80fb12116914e533e5f11ecf26f337751abc.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4277,25 +4277,25 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX max.f16x2 with =r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4070d164dfb7ec86737afcc51fdef8e9452193df799721221b988492eab3188."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined three-form sm_80 packed-f16x2 upper probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9dc936add8025c51ff86d52cb72929dbbd1197519c0ffed33b53ba5eae61c7c.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4329,25 +4329,25 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX neg.f16x2 with =r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact neg.f16x2 inline PTX at PTX 6.0/sm_70. An independent generic v2f16 selection probe emitted neg.f16x2 at sm_53; at sm_52 LLVM emitted scalar sign-bit operations and no packed neg instruction."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e44e1921be6e1f3bc1cc0e885622b5bb8a552876f0f300d6cff83701c2b0a8."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "PTX 6.0 emitted by the sm_70 neg.f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 737-byte PTX module at sm_75 and produced a 3,112-byte cubin (SHA-256 0e381d181a5d707df9d2437d636d47bf1f80cdff578b38160e3553bec0bc870a). SASS realizes packed negation as HADD2 with negated half lanes. The PTX 5.0 negative module was rejected because neg.f16/neg.f16x2 requires PTX 6.0 or later.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6abb77861c39cdcaed05cd8512d1b2ca13a32c045cecedb72aa4038fc9c0d6e. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -4397,25 +4397,25 @@
             "sm_70",
             "ptx65"
           ],
-          "representation": "pure inline PTX abs.f16x2 with =r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact abs.f16x2 inline PTX at PTX 6.5/sm_70. An independent typed call to the resolved llvm.nvvm.fabs.v2f16 symbol emitted the same instruction at PTX 6.5/sm_70 and at its native sm_53 floor; sm_52 failed exact intrinsic selection."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7282ee206f4fdb3a54d1fe47412d9ab41a11954358a8e0ee203f434988aa5191."
         },
         {
           "targets": [
             "sm_75",
             "ptx65"
           ],
-          "representation": "PTX 6.5 emitted by the sm_70 abs.f16x2 probe and assembled at CUDA 13.3's supported sm_75 floor",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 737-byte PTX module at sm_75 and produced a 3,112-byte cubin (SHA-256 24150b058e7b7ad99ffeacd77588a2f1febdf5479f93f966ac0192c9834babf7). SASS realizes packed absolute value as HADD2 with an absolute source. The PTX 6.4 negative module was rejected because abs.f16/abs.f16x2 requires PTX 6.5 or later.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93adc93cea87ea659d8a0b4542f44e1d897cb4c1cd53a06dc08becdac7de7ea6.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true,
@@ -4482,62 +4482,62 @@
           "targets": [
             "sm_80"
           ],
-          "representation": "typed intrinsic llvm.nvvm.ff2f16x2.rn(float, float) -> <2 x half>",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60)."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 23c48425228bb21090b442cbcb63b58874efaf6cc05dde2ad7549aff23e3f8ce."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed call to llvm.nvvm.ff2f16x2.rn(float high, float low) -> <2 x half>",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7c94821ab335a6f02f2b3ca52d1e0c2428d1d26aac5e7c62852ef35cbfcba8e4."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d.",
           "artifact_kind": "cubin",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX cvt.rn.f16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rn, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 009a800094f29371a5f120862eb39d45b097aaa5715ce2e7ac26a43ec3665773."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true
@@ -4589,62 +4589,62 @@
           "targets": [
             "sm_80"
           ],
-          "representation": "typed intrinsic llvm.nvvm.ff2f16x2.rz(float, float) -> <2 x half>",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60)."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2bccf7085ce3ccc76e323145e0cd1e86fda391d0411adc51c1eea4ce806a2f03."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed call to llvm.nvvm.ff2f16x2.rz(float high, float low) -> <2 x half>",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rz.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5790119795c6ad9cbfc37ff239e9a123b44c05bde5d38fc5675565933d23a6ff."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e.",
           "artifact_kind": "cubin",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX cvt.rz.f16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rz.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rz, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6833e6724886aa52db6de2c812bc6e0e2a7216e33ba07501701559eb4c68cf5c."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true
@@ -4697,62 +4697,62 @@
           "targets": [
             "sm_80"
           ],
-          "representation": "typed intrinsic llvm.nvvm.ff2f16x2.rn.relu(float, float) -> <2 x half>",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60)."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f4b2d11b6f47357e52b5da2bd8374d6043945e2d3c7d28f0d9bce741658cbbfb."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed call to llvm.nvvm.ff2f16x2.rn.relu(float high, float low) -> <2 x half>",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.relu.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a886266d7776914c367d748d72bd9b1032db99c5eb5366f902a3b5b8e63f2add."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8.",
           "artifact_kind": "cubin",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX cvt.rn.relu.f16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.relu.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rn.relu, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 644949d91a98e2c1b36b125ca7211d181b486d46e6b0335c823f0f78de9860cc."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true
@@ -4805,62 +4805,62 @@
           "targets": [
             "sm_80"
           ],
-          "representation": "typed intrinsic llvm.nvvm.ff2bf16x2.rn.relu(float, float) -> <2 x bfloat>",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60)."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e02262263ec043530dec4469b22cfd6e831bc597d97b59e20132167985ae8e75."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed call to llvm.nvvm.ff2bf16x2.rn.relu(float high, float low) -> <2 x bfloat>",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.relu.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c797710407ad5b5c6425fc3ae32587456573e2442fb1398a784e61ab025c7f6b."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806.",
           "artifact_kind": "cubin",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX cvt.rn.relu.bf16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.relu.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rn.relu, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 336ac20640c8b6451d754f6afed9479d89a22cd5c47b6606e57f251337b35e0c."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true
@@ -4912,62 +4912,62 @@
           "targets": [
             "sm_80"
           ],
-          "representation": "typed intrinsic llvm.nvvm.ff2bf16x2.rz(float, float) -> <2 x bfloat>",
+          "representation": "separate typed LLVM declaration probes",
           "stage": "declaration_canonicalization",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60)."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb6bcfe8642c6981459470ff9b577231bb7c5575846df5a6b524e3ed834bfc4c."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "typed call to llvm.nvvm.ff2bf16x2.rz(float high, float low) -> <2 x bfloat>",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend emitted exact cvt.rz.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dcafda6629d6de7301da44bf1e7690a5d1e8131d595cc975ea4e8b68a61de5d."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX 7.0 emitted by the combined six-form typed LLVM probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "typed_nvvm",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6.",
           "artifact_kind": "cubin",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX cvt.rz.bf16x2.f32 with `$0, $2, $1` and `=r,f,f`",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The selected Rust LLVM 22.1.2 backend retained exact cvt.rz.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rz, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd97b8bfbf98ce29b538e25417e266f1d5c29d1d18b0789af1dd439fe4431d90."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "PTX emitted by the combined six-form sm_80 packed-conversion probe",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "declaration_attributes_canonicalized": true
@@ -4994,26 +4994,26 @@
             "sm_20",
             "ptx32"
           ],
-          "representation": "side-effect inline PTX trap in the combined debug-control probe",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted PTX 3.2 containing trap; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32b430781d5a0e49a3cf431871f05da67c5f0a37e984050e1a2030c9f4e4f283."
         },
         {
           "targets": [
             "sm_75",
             "ptx32"
           ],
-          "representation": "LLVM-produced PTX module containing trap",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the combined debug-control module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b62e57d3cf50fcc7ced943d8501901809cc7d321cab2fbfb101791ed0909718. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -5039,26 +5039,26 @@
             "sm_20",
             "ptx32"
           ],
-          "representation": "side-effect inline PTX brkpt in the combined debug-control probe",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted PTX 3.2 containing brkpt; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 612fb1dc878abe6a363a52b221b9b188b38d5707df5567d98e8e0c26bd0902b4."
         },
         {
           "targets": [
             "sm_75",
             "ptx32"
           ],
-          "representation": "LLVM-produced PTX module containing brkpt",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the combined debug-control module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 66e4eb480c47c5102813d6f0b276ec783bf14a6f3bdc849b69eca9c086bfed5c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -5088,26 +5088,26 @@
             "sm_20",
             "ptx32"
           ],
-          "representation": "side-effect inline PTX pmevent 15 in the combined debug-control probe",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "The pinned Rust LLVM emitted PTX 3.2 containing pmevent 15; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6af969aa9d4c327529700e8a82bc7e08ce798613e609091e74e47a57adaf113f."
         },
         {
           "targets": [
             "sm_75",
             "ptx32"
           ],
-          "representation": "LLVM-produced PTX module containing pmevent 15",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the combined debug-control module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7c77d308d6af079d03e08c447e632722a4dbb282c7d6faea32b7e1174e9f623. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -5146,26 +5146,26 @@
             "sm_75",
             "ptx78"
           ],
-          "representation": "one convergent LLVM IR inline-PTX probe for the packed b16 transpose",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned Rust LLVM emitted one exact movmatrix instruction; source SHA-256 b8776ed95d8c8e451deea02d8851a238cfcfe570e286a92c94e5971bf06db963, bitcode SHA-256 caf25ab44ed893d449343707f06af85395e86f41852c32ec1b577be184365aee, and PTX SHA-256 f1e1b7d8ad879abd120d6cb32a5999925af0ed77b3a7144a7dfd8db76ea18246."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4d774d39425c57263a42ec917b671796b75c2e755a81255dfc72a40232e32367."
         },
         {
           "targets": [
             "sm_75",
             "ptx78"
           ],
-          "representation": "642-byte LLVM-produced PTX module containing movmatrix.sync.aligned.m8n8.trans.b16",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 85c6f2304703e7ee71bd8ff7bbb2dec8730ea51244c37934a0cc838b2d00ab69.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ]
     },
@@ -6764,26 +6764,26 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX fma.rn.ftz.f16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80a7eb25c335cf849bc1e0be9ed1e391b9d311a1573d38be927d6aa35c8b4aa3."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83de32412117576acec7391b12d4c1e5f2f026f4fed9281689ec4a87e9b1deaf. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -6832,26 +6832,26 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX fma.rn.sat.f16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7ccf304fffd63a678dec5f5775eb1fbdc60e0a295b7b2a9e878339cf8a0922a."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ad17848cb5677cf0a68e6e3eb435b41a165f6ef1cbaf69b0fa3e2b4d8bbfdcfe. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -6900,26 +6900,26 @@
             "sm_70",
             "ptx60"
           ],
-          "representation": "pure inline PTX fma.rn.ftz.sat.f16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9bed831d23f91ca28a7fa178103ea679a4765ec587f918986874e52b21d00da1."
         },
         {
           "targets": [
             "sm_75",
             "ptx60"
           ],
-          "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms.",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6a185d53a7cf04592fc205ee4ef0926f252fa57f5c50f1ee510c7aa6210578ee. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
@@ -6969,26 +6969,26 @@
             "sm_80",
             "ptx70"
           ],
-          "representation": "pure inline PTX fma.rn.ftz.relu.f16x2 with =r,r,r,r",
+          "representation": "separate probes for the recorded lowering mechanism",
           "stage": "backend_codegen",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
-          "detail": "Pinned llc retained all 1 exact instruction head(s) in 682-byte PTX (SHA-256 ee283e72de7ddec2258b4eb628ca180889b67d9f79d13ba926719724bec1c01f) lowered from the 196-byte fixture (SHA-256 fa242c0f862cc57936c53949b30d4ce59871e59cde0349570a44122c3a164006). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads."
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93c9c1a8aaa9ac2bee924f4b1eee264e0aa38c8c6f01c04bddf3bfd58d169bc4."
         },
         {
           "targets": [
             "sm_80",
             "ptx70"
           ],
-          "representation": "inline PTX assembled by the pinned CUDA 13.3 ptxas",
+          "representation": "separate PTX outputs assembled for the recorded target",
           "stage": "ptx_assembly",
           "mechanism": "inline_ptx",
           "outcome": "succeeded",
           "artifact_kind": "cubin",
-          "detail": "CUDA 13.3 ptxas accepted the module at sm_80 and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2).",
+          "detail": "Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 582782aeec6a6286313dd28330e3de2a23eeaa07c479ee6271c831e0c8f9d1e7.",
           "tool_path": "/usr/local/cuda-13.3/bin/ptxas",
-          "tool_version": "CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0",
-          "tool_sha256": "7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb"
+          "tool_version": "Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0",
+          "tool_sha256": "afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc"
         }
       ],
       "runtime_validation": "unexecuted",
diff --git a/cuda-oxide/intrinsics/generated-reference.md b/cuda-oxide/intrinsics/generated-reference.md
index 71c68648ee..44c2b00700 100644
--- a/cuda-oxide/intrinsics/generated-reference.md
+++ b/cuda-oxide/intrinsics/generated-reference.md
@@ -2,8 +2,8 @@
 SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 SPDX-License-Identifier: Apache-2.0
 @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 -->
 
 # Generated Intrinsic Reference
@@ -12,1035 +12,1035 @@ This table is generated from the resolved catalog. Evidence stages below disting
 
 | Rust function | CUDA operation | Source | Effects | Availability | Backend evidence |
 |:--|:--|:--|:--|:--|:--|
-| `cuda_intrinsics::bf16x2::abs_bf16x2` | `nvvm.abs_bf16x2` | LLVM `llvm.nvvm.fabs` from `int_nvvm_fabs` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.6 Half Precision Floating Point Instructions: abs](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs)) | validated on `rust-llvm-23.1.0-16696adc`; expects `abs.bf16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::abs_f16x2` | `nvvm.abs_f16x2` | LLVM `llvm.nvvm.fabs` from `int_nvvm_fabs` | safe; scope thread; pure; memory none; convergent false | PTX 6.5 on sm_70+ ([PTX ISA 9.3, 9.7.4.6 Half Precision Floating Point Instructions: abs](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs)) | validated on `rust-llvm-23.1.0-16696adc`; expects `abs.f16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::active_mask` | `nvvm.activemask` | LLVM `llvm.nvvm.activemask` from `int_nvvm_activemask` | safe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.2 on sm_30+ ([PTX ISA 9.3, 9.7.14.12 Parallel Synchronization and Communication Instructions: activemask](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-activemask)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `activemask.b32 <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::add_bf16x2` | `nvvm.add_bf16x2` | PTX-native `add.rn.bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.4.1 Half Precision Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::add_f16x2` | `nvvm.add_f16x2` | PTX-native `add.rn.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.1 Half Precision Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::add_f32x2` | `nvvm.add_f32x2` | PTX-native `add.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.1 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::add_ftz_f32x2` | `nvvm.add_ftz_f32x2` | PTX-native `add.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.1 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.ftz.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.f` from `int_nvvm_add_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rm.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.d` from `int_nvvm_add_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rm.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.ftz.f` from `int_nvvm_add_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rm.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rm_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.ftz.sat.f` from `int_nvvm_add_rm_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rm.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rm_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.sat.f` from `int_nvvm_add_rm_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rm.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.f` from `int_nvvm_add_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.d` from `int_nvvm_add_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.ftz.f` from `int_nvvm_add_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rn_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.ftz.sat.f` from `int_nvvm_add_rn_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rn_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.sat.f` from `int_nvvm_add_rn_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rn.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.f` from `int_nvvm_add_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rp.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.d` from `int_nvvm_add_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rp.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.ftz.f` from `int_nvvm_add_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rp.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rp_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.ftz.sat.f` from `int_nvvm_add_rp_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rp.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rp_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.sat.f` from `int_nvvm_add_rp_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rp.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.f` from `int_nvvm_add_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.d` from `int_nvvm_add_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rz.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.ftz.f` from `int_nvvm_add_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rz.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rz_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.ftz.sat.f` from `int_nvvm_add_rz_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rz.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::add_rz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.sat.f` from `int_nvvm_add_rz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.0-16696adc`; expects `add.rz.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::all_sync` | `nvvm.vote_sync_all` | LLVM `llvm.nvvm.vote.all.sync` from `int_nvvm_vote_all_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `vote.sync.all.pred <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::any_sync` | `nvvm.vote_sync_any` | LLVM `llvm.nvvm.vote.any.sync` from `int_nvvm_vote_any_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `vote.sync.any.pred <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::ballot_sync` | `nvvm.vote_sync_ballot` | LLVM `llvm.nvvm.vote.ballot.sync` from `int_nvvm_vote_ballot_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `vote.sync.ballot.b32 <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::barrier_cluster_arrive` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive` from `int_nvvm_barrier_cluster_arrive` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.0-16696adc`; expects `barrier.cluster.arrive;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::barrier_cluster_arrive_aligned` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive.aligned` from `int_nvvm_barrier_cluster_arrive_aligned` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.0-16696adc`; expects `barrier.cluster.arrive.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::barrier_cluster_arrive_relaxed` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive.relaxed` from `int_nvvm_barrier_cluster_arrive_relaxed` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.0-16696adc`; expects `barrier.cluster.arrive.relaxed;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::barrier_cluster_arrive_relaxed_aligned` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive.relaxed.aligned` from `int_nvvm_barrier_cluster_arrive_relaxed_aligned` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.0-16696adc`; expects `barrier.cluster.arrive.relaxed.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::barrier_cluster_wait` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.wait` from `int_nvvm_barrier_cluster_wait` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.0-16696adc`; expects `barrier.cluster.wait;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::barrier_cluster_wait_aligned` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.wait.aligned` from `int_nvvm_barrier_cluster_wait_aligned` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.0-16696adc`; expects `barrier.cluster.wait.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::barrier_cta_arrive` | `nvvm.barrier_cta_arrive_count` | LLVM `llvm.nvvm.barrier.cta.arrive.count` from `int_nvvm_barrier_cta_arrive_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `barrier.arrive <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::barrier_cta_arrive_aligned` | `nvvm.barrier_cta_arrive_aligned_count` | LLVM `llvm.nvvm.barrier.cta.arrive.aligned.count` from `int_nvvm_barrier_cta_arrive_aligned_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `bar.arrive <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::barrier_cta_sync` | `nvvm.barrier_cta_sync_count` | LLVM `llvm.nvvm.barrier.cta.sync.count` from `int_nvvm_barrier_cta_sync_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `barrier.sync <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::barrier_cta_sync_aligned` | `nvvm.barrier_cta_sync_aligned_count` | LLVM `llvm.nvvm.barrier.cta.sync.aligned.count` from `int_nvvm_barrier_cta_sync_aligned_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `bar.sync <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::block_dim_x` | `nvvm.read_ptx_sreg_ntid_x` | LLVM `llvm.nvvm.read.ptx.sreg.ntid.x` from `int_nvvm_read_ptx_sreg_ntid_x` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.2 Special Registers: %ntid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %ntid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::block_dim_y` | `nvvm.read_ptx_sreg_ntid_y` | LLVM `llvm.nvvm.read.ptx.sreg.ntid.y` from `int_nvvm_read_ptx_sreg_ntid_y` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.2 Special Registers: %ntid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %ntid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::block_dim_z` | `nvvm.read_ptx_sreg_ntid_z` | LLVM `llvm.nvvm.read.ptx.sreg.ntid.z` from `int_nvvm_read_ptx_sreg_ntid_z` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.2 Special Registers: %ntid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %ntid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::block_idx_x` | `nvvm.read_ptx_sreg_ctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.ctaid.x` from `int_nvvm_read_ptx_sreg_ctaid_x` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.6 Special Registers: %ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %ctaid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::block_idx_y` | `nvvm.read_ptx_sreg_ctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.ctaid.y` from `int_nvvm_read_ptx_sreg_ctaid_y` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.6 Special Registers: %ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %ctaid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::block_idx_z` | `nvvm.read_ptx_sreg_ctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.ctaid.z` from `int_nvvm_read_ptx_sreg_ctaid_z` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.6 Special Registers: %ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %ctaid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::debug::breakpoint` | `nvvm.brkpt` | PTX-native `brkpt` | safe; scope thread; impure; memory none; convergent false | PTX 1.0 on sm_11+ ([PTX ISA 9.3, 9.7.20.1 Miscellaneous Instructions: brkpt](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-brkpt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `brkpt;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::clc::clc_query_get_first_ctaid_x` | `nvvm.clc_query_get_first_ctaid_x` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.get_first_ctaid.x` from `int_nvvm_clusterlaunchcontrol_query_cancel_get_first_ctaid_x` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.0-16696adc`; expects `clusterlaunchcontrol.query_cancel.get_first_ctaid::x.b32.b128 <register>, %clc_handle;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::clc::clc_query_get_first_ctaid_y` | `nvvm.clc_query_get_first_ctaid_y` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.get_first_ctaid.y` from `int_nvvm_clusterlaunchcontrol_query_cancel_get_first_ctaid_y` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.0-16696adc`; expects `clusterlaunchcontrol.query_cancel.get_first_ctaid::y.b32.b128 <register>, %clc_handle;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::clc::clc_query_get_first_ctaid_z` | `nvvm.clc_query_get_first_ctaid_z` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.get_first_ctaid.z` from `int_nvvm_clusterlaunchcontrol_query_cancel_get_first_ctaid_z` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.0-16696adc`; expects `clusterlaunchcontrol.query_cancel.get_first_ctaid::z.b32.b128 <register>, %clc_handle;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::clc::clc_query_is_canceled` | `nvvm.clc_query_is_canceled` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.is_canceled` from `int_nvvm_clusterlaunchcontrol_query_cancel_is_canceled` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.0-16696adc`; expects `clusterlaunchcontrol.query_cancel.is_canceled.pred.b128 <register>, %clc_handle;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::clc::clc_try_cancel` | `nvvm.clc_try_cancel` | LLVM `llvm.nvvm.clusterlaunchcontrol.try_cancel.async.shared` from `int_nvvm_clusterlaunchcontrol_try_cancel_async_shared` | unsafe; scope cta; impure; memory read_write; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.0-16696adc`; expects `clusterlaunchcontrol.try_cancel.async.shared::cta.mbarrier::complete_tx::bytes.b128 [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::clc::clc_try_cancel_multicast` | `nvvm.clc_try_cancel_multicast` | LLVM `llvm.nvvm.clusterlaunchcontrol.try_cancel.async.multicast.shared` from `int_nvvm_clusterlaunchcontrol_try_cancel_async_multicast_shared` | unsafe; scope cta; impure; memory read_write; convergent false | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a or sm_120a or sm_121a ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.0-16696adc`; expects `clusterlaunchcontrol.try_cancel.async.shared::cta.mbarrier::complete_tx::bytes.multicast::cluster::all.b128 [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::debug::clock` | `nvvm.read_ptx_sreg_clock` | LLVM `llvm.nvvm.read.ptx.sreg.clock` from `int_nvvm_read_ptx_sreg_clock` | safe; scope sm; impure; memory inaccessible_read_write; convergent false | PTX 1.0 on all ([PTX ISA 9.3, 10.23 Special Registers: %clock, %clock_hi](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clock-clock-hi)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %clock;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::debug::clock64` | `nvvm.read_ptx_sreg_clock64` | LLVM `llvm.nvvm.read.ptx.sreg.clock64` from `int_nvvm_read_ptx_sreg_clock64` | safe; scope sm; impure; memory inaccessible_read_write; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.24 Special Registers: %clock64](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clock64)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u64 <register>, %clock64;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_block_count` | `nvvm.read_ptx_sreg_cluster_nctarank` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctarank` from `int_nvvm_read_ptx_sreg_cluster_nctarank` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.17 Special Registers: %cluster_nctarank](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctarank)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_nctarank;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_block_idx_x` | `nvvm.read_ptx_sreg_cluster_ctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctaid.x` from `int_nvvm_read_ptx_sreg_cluster_ctaid_x` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.14 Special Registers: %cluster_ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_ctaid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_block_idx_y` | `nvvm.read_ptx_sreg_cluster_ctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctaid.y` from `int_nvvm_read_ptx_sreg_cluster_ctaid_y` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.14 Special Registers: %cluster_ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_ctaid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_block_idx_z` | `nvvm.read_ptx_sreg_cluster_ctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctaid.z` from `int_nvvm_read_ptx_sreg_cluster_ctaid_z` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.14 Special Registers: %cluster_ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_ctaid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_block_rank` | `nvvm.read_ptx_sreg_cluster_ctarank` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctarank` from `int_nvvm_read_ptx_sreg_cluster_ctarank` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.16 Special Registers: %cluster_ctarank](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctarank)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_ctarank;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_dim_x` | `nvvm.read_ptx_sreg_cluster_nctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctaid.x` from `int_nvvm_read_ptx_sreg_cluster_nctaid_x` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.15 Special Registers: %cluster_nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_nctaid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_dim_y` | `nvvm.read_ptx_sreg_cluster_nctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctaid.y` from `int_nvvm_read_ptx_sreg_cluster_nctaid_y` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.15 Special Registers: %cluster_nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_nctaid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_dim_z` | `nvvm.read_ptx_sreg_cluster_nctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctaid.z` from `int_nvvm_read_ptx_sreg_cluster_nctaid_z` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.15 Special Registers: %cluster_nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %cluster_nctaid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_grid_dim_x` | `nvvm.read_ptx_sreg_nclusterid_x` | LLVM `llvm.nvvm.read.ptx.sreg.nclusterid.x` from `int_nvvm_read_ptx_sreg_nclusterid_x` | safe; scope grid; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.13 Special Registers: %nclusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nclusterid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_grid_dim_y` | `nvvm.read_ptx_sreg_nclusterid_y` | LLVM `llvm.nvvm.read.ptx.sreg.nclusterid.y` from `int_nvvm_read_ptx_sreg_nclusterid_y` | safe; scope grid; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.13 Special Registers: %nclusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nclusterid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_grid_dim_z` | `nvvm.read_ptx_sreg_nclusterid_z` | LLVM `llvm.nvvm.read.ptx.sreg.nclusterid.z` from `int_nvvm_read_ptx_sreg_nclusterid_z` | safe; scope grid; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.13 Special Registers: %nclusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nclusterid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_idx_x` | `nvvm.read_ptx_sreg_clusterid_x` | LLVM `llvm.nvvm.read.ptx.sreg.clusterid.x` from `int_nvvm_read_ptx_sreg_clusterid_x` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.12 Special Registers: %clusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %clusterid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_idx_y` | `nvvm.read_ptx_sreg_clusterid_y` | LLVM `llvm.nvvm.read.ptx.sreg.clusterid.y` from `int_nvvm_read_ptx_sreg_clusterid_y` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.12 Special Registers: %clusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %clusterid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::cluster_idx_z` | `nvvm.read_ptx_sreg_clusterid_z` | LLVM `llvm.nvvm.read.ptx.sreg.clusterid.z` from `int_nvvm_read_ptx_sreg_clusterid_z` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.12 Special Registers: %clusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %clusterid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::cos_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.cos.approx.f` from `int_nvvm_cos_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.11 Floating Point Instructions: cos](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-cos)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cos.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::cos_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.cos.approx.ftz.f` from `int_nvvm_cos_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.11 Floating Point Instructions: cos](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-cos)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cos.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_commit_group` | `nvvm.cp_async_bulk_commit_group` | LLVM `llvm.nvvm.cp.async.bulk.commit.group` from `int_nvvm_cp_async_bulk_commit_group` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.commit_group;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_1d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_1d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_1d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.1d.L2.global.tile [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_1d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_1d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_1d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.1d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_2d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_2d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_2d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_2d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_3d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_3d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_3d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.3d.L2.global.tile [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_3d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_3d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_3d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.3d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_4d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_4d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_4d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.4d.L2.global.tile [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_4d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_4d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_4d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.4d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_5d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_5d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_5d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.5d.L2.global.tile [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_5d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_5d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_5d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.5d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_gather4_2d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_gather4_2d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.gather4.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_gather4_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile::gather4 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.gather4.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_gather4_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile::gather4.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_1d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.1d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.1d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_1d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.1d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.1d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.2d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_g2s_multicast` | `nvvm.cp_async_bulk_tensor_g2s_tile_2d_multicast` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.2d.shared::cluster.global.tile.mbarrier::complete_tx::bytes.multicast::cluster [<address>], [<address>], [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_g2s_multicast_cg2` | `nvvm.cp_async_bulk_tensor_g2s_tile_2d_multicast_cg2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.2d.shared::cluster.global.tile.mbarrier::complete_tx::bytes.multicast::cluster.cta_group::2 [<address>], [<address>], [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.2d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.2d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_3d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.3d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.3d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_3d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.3d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.3d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_4d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.4d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.4d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_4d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.4d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.4d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_5d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.5d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.5d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_5d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.5d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.tensor.5d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_add_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.add.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_add_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.add.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_add_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.add.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_and_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.and.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_and_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.and.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_and_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.and.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_dec_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.dec.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_dec_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.dec.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_dec_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.dec.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_inc_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.inc.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_inc_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.inc.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_inc_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.inc.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_max_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.max.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_max_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.max.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_max_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.max.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_min_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.min.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_min_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.min.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_min_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.min.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_or_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.or.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_or_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.or.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_or_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.or.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_xor_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.xor.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_xor_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.xor.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_xor_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.xor.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_wait_group` | `nvvm.cp_async_bulk_wait_group` | LLVM `llvm.nvvm.cp.async.bulk.wait.group` from `int_nvvm_cp_async_bulk_wait_group` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.wait_group <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::cp_async_bulk_wait_group_read` | `nvvm.cp_async_bulk_wait_group_read` | LLVM `llvm.nvvm.cp.async.bulk.wait.group.read` from `int_nvvm_cp_async_bulk_wait_group_read` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `cp.async.bulk.wait_group.read <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_ca_16` | `nvvm.cp_async_ca_16` | LLVM `llvm.nvvm.cp.async.ca.shared.global.16` from `int_nvvm_cp_async_ca_shared_global_16` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.ca.shared.global [<address>], [<address>], 16;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_ca_4` | `nvvm.cp_async_ca_4` | LLVM `llvm.nvvm.cp.async.ca.shared.global.4` from `int_nvvm_cp_async_ca_shared_global_4` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.ca.shared.global [<address>], [<address>], 4;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_ca_8` | `nvvm.cp_async_ca_8` | LLVM `llvm.nvvm.cp.async.ca.shared.global.8` from `int_nvvm_cp_async_ca_shared_global_8` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.ca.shared.global [<address>], [<address>], 8;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_ca_zfill_16` | `nvvm.cp_async_ca_zfill_16` | LLVM `llvm.nvvm.cp.async.ca.shared.global.16.s` from `int_nvvm_cp_async_ca_shared_global_16_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.ca.shared.global [<address>], [<address>], 16, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_ca_zfill_4` | `nvvm.cp_async_ca_zfill_4` | LLVM `llvm.nvvm.cp.async.ca.shared.global.4.s` from `int_nvvm_cp_async_ca_shared_global_4_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.ca.shared.global [<address>], [<address>], 4, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_ca_zfill_8` | `nvvm.cp_async_ca_zfill_8` | LLVM `llvm.nvvm.cp.async.ca.shared.global.8.s` from `int_nvvm_cp_async_ca_shared_global_8_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.ca.shared.global [<address>], [<address>], 8, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_cg_16` | `nvvm.cp_async_cg_16` | LLVM `llvm.nvvm.cp.async.cg.shared.global.16` from `int_nvvm_cp_async_cg_shared_global_16` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.cg.shared.global [<address>], [<address>], 16;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_cg_zfill_16` | `nvvm.cp_async_cg_zfill_16` | LLVM `llvm.nvvm.cp.async.cg.shared.global.16.s` from `int_nvvm_cp_async_cg_shared_global_16_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.cg.shared.global [<address>], [<address>], 16, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_commit_group` | `nvvm.cp_async_commit_group` | LLVM `llvm.nvvm.cp.async.commit.group` from `int_nvvm_cp_async_commit_group` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.2 Data Movement and Conversion Instructions: cp.async.commit_group](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-commit-group)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.commit_group;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive` | `nvvm.cp_async_mbarrier_arrive` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive` from `int_nvvm_cp_async_mbarrier_arrive` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.mbarrier.arrive.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive_noinc` | `nvvm.cp_async_mbarrier_arrive_noinc` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive.noinc` from `int_nvvm_cp_async_mbarrier_arrive_noinc` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.mbarrier.arrive.noinc.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive_noinc_shared` | `nvvm.cp_async_mbarrier_arrive_noinc_shared` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive.noinc.shared` from `int_nvvm_cp_async_mbarrier_arrive_noinc_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.mbarrier.arrive.noinc.shared.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive_shared` | `nvvm.cp_async_mbarrier_arrive_shared` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive.shared` from `int_nvvm_cp_async_mbarrier_arrive_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.mbarrier.arrive.shared.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_wait_all` | `nvvm.cp_async_wait_all` | LLVM `llvm.nvvm.cp.async.wait.all` from `int_nvvm_cp_async_wait_all` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.3 Data Movement and Conversion Instructions: cp.async.wait_group / cp.async.wait_all](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.wait_all;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::async_copy::cp_async_wait_group` | `nvvm.cp_async_wait_group` | LLVM `llvm.nvvm.cp.async.wait.group` from `int_nvvm_cp_async_wait_group` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.3 Data Movement and Conversion Instructions: cp.async.wait_group / cp.async.wait_all](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cp.async.wait_group <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_f16x2_f32` | `nvvm.cvt_f16x2_f32` | LLVM `llvm.nvvm.ff2f16x2.rn` from `int_nvvm_ff2f16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.f16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_f32x2_bf16x2` | `nvvm.cvt_f32x2_bf16x2` | LLVM `llvm.nvvm.ff2bf16x2.rn` from `int_nvvm_ff2bf16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.bf16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_f16x2_e4m3x2` | `nvvm.cvt_rn_f16x2_e4m3x2` | LLVM `llvm.nvvm.e4m3x2.to.f16x2.rn` from `int_nvvm_e4m3x2_to_f16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.f16x2.e4m3x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_f16x2_e5m2x2` | `nvvm.cvt_rn_f16x2_e5m2x2` | LLVM `llvm.nvvm.e5m2x2.to.f16x2.rn` from `int_nvvm_e5m2x2_to_f16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.f16x2.e5m2x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_relu_bf16x2_f32` | `nvvm.cvt_rn_relu_bf16x2_f32` | LLVM `llvm.nvvm.ff2bf16x2.rn.relu` from `int_nvvm_ff2bf16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.relu.bf16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_relu_f16x2_e4m3x2` | `nvvm.cvt_rn_relu_f16x2_e4m3x2` | LLVM `llvm.nvvm.e4m3x2.to.f16x2.rn.relu` from `int_nvvm_e4m3x2_to_f16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.relu.f16x2.e4m3x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_relu_f16x2_e5m2x2` | `nvvm.cvt_rn_relu_f16x2_e5m2x2` | LLVM `llvm.nvvm.e5m2x2.to.f16x2.rn.relu` from `int_nvvm_e5m2x2_to_f16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.relu.f16x2.e5m2x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_relu_f16x2_f32` | `nvvm.cvt_rn_relu_f16x2_f32` | LLVM `llvm.nvvm.ff2f16x2.rn.relu` from `int_nvvm_ff2f16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.relu.f16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_relu_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn.relu.satfinite` from `int_nvvm_f2tf32_rn_relu_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.relu.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_relu_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn.relu` from `int_nvvm_f2tf32_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.relu.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_e4m3x2_f16x2` | `nvvm.cvt_rn_satfinite_e4m3x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e4m3x2.rn` from `int_nvvm_f16x2_to_e4m3x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.e4m3x2.f16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_e4m3x2_f32` | `nvvm.cvt_rn_satfinite_e4m3x2_f32` | LLVM `llvm.nvvm.ff.to.e4m3x2.rn` from `int_nvvm_ff_to_e4m3x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.e4m3x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_e5m2x2_f16x2` | `nvvm.cvt_rn_satfinite_e5m2x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e5m2x2.rn` from `int_nvvm_f16x2_to_e5m2x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.e5m2x2.f16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_e5m2x2_f32` | `nvvm.cvt_rn_satfinite_e5m2x2_f32` | LLVM `llvm.nvvm.ff.to.e5m2x2.rn` from `int_nvvm_ff_to_e5m2x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.e5m2x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e4m3x2_f16x2` | `nvvm.cvt_rn_satfinite_relu_e4m3x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e4m3x2.rn.relu` from `int_nvvm_f16x2_to_e4m3x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.relu.e4m3x2.f16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e4m3x2_f32` | `nvvm.cvt_rn_satfinite_relu_e4m3x2_f32` | LLVM `llvm.nvvm.ff.to.e4m3x2.rn.relu` from `int_nvvm_ff_to_e4m3x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.relu.e4m3x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e5m2x2_f16x2` | `nvvm.cvt_rn_satfinite_relu_e5m2x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e5m2x2.rn.relu` from `int_nvvm_f16x2_to_e5m2x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.relu.e5m2x2.f16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e5m2x2_f32` | `nvvm.cvt_rn_satfinite_relu_e5m2x2_f32` | LLVM `llvm.nvvm.ff.to.e5m2x2.rn.relu` from `int_nvvm_ff_to_e5m2x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.relu.e5m2x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn.satfinite` from `int_nvvm_f2tf32_rn_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rn_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn` from `int_nvvm_f2tf32_rn` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rn.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rna_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rna.satfinite` from `int_nvvm_f2tf32_rna_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rna.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rna_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rna` from `int_nvvm_f2tf32_rna` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rna.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rz_bf16x2_f32` | `nvvm.cvt_rz_bf16x2_f32` | LLVM `llvm.nvvm.ff2bf16x2.rz` from `int_nvvm_ff2bf16x2_rz` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rz.bf16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rz_f16x2_f32` | `nvvm.cvt_rz_f16x2_f32` | LLVM `llvm.nvvm.ff2f16x2.rz` from `int_nvvm_ff2f16x2_rz` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rz.f16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rz_relu_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz.relu.satfinite` from `int_nvvm_f2tf32_rz_relu_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rz.relu.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rz_relu_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz.relu` from `int_nvvm_f2tf32_rz_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rz.relu.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rz_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz.satfinite` from `int_nvvm_f2tf32_rz_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rz.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::convert::cvt_rz_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz` from `int_nvvm_f2tf32_rz` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `cvt.rz.tf32.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rm.f` from `int_nvvm_div_rm_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rm.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rm.d` from `int_nvvm_div_rm_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rm.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rm.ftz.f` from `int_nvvm_div_rm_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rm.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rn.f` from `int_nvvm_div_rn_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rn.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rn.d` from `int_nvvm_div_rn_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rn.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rn.ftz.f` from `int_nvvm_div_rn_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rn.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rp.f` from `int_nvvm_div_rp_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rp.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rp.d` from `int_nvvm_div_rp_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rp.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rp.ftz.f` from `int_nvvm_div_rp_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rp.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rz.f` from `int_nvvm_div_rz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rz.d` from `int_nvvm_div_rz_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rz.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::div_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rz.ftz.f` from `int_nvvm_div_rz_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.0-16696adc`; expects `div.rz.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::dotprod::dp2a_s32` | `nvvm.dp2a_s32` | LLVM `llvm.nvvm.idp2a.s.s` from `int_nvvm_idp2a_s_s` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.25 Integer Arithmetic Instructions: dp2a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a)) | validated on `rust-llvm-23.1.0-16696adc`; expects `dp2a.lo.s32.s32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::dotprod::dp2a_u32` | `nvvm.dp2a_u32` | LLVM `llvm.nvvm.idp2a.u.u` from `int_nvvm_idp2a_u_u` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.25 Integer Arithmetic Instructions: dp2a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a)) | validated on `rust-llvm-23.1.0-16696adc`; expects `dp2a.lo.u32.u32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::dotprod::dp4a_s32` | `nvvm.dp4a_s32` | LLVM `llvm.nvvm.idp4a.s.s` from `int_nvvm_idp4a_s_s` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.24 Integer Arithmetic Instructions: dp4a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a)) | validated on `rust-llvm-23.1.0-16696adc`; expects `dp4a.s32.s32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::dotprod::dp4a_u32` | `nvvm.dp4a_u32` | LLVM `llvm.nvvm.idp4a.u.u` from `int_nvvm_idp4a_u_u` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.24 Integer Arithmetic Instructions: dp4a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a)) | validated on `rust-llvm-23.1.0-16696adc`; expects `dp4a.u32.u32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::dsmem_read_u32` | `nvvm.dsmem_read_u32` | PTX-native `mapa.shared::cluster.u64 + ld.shared::cluster.u32` | unsafe; scope cluster; impure; memory read; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.8 Data Movement and Conversion Instructions: ld](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-ld)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ld.shared::cluster.u32 <register>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::shared::dynamic_smem_size` | `nvvm.read_ptx_sreg_dynamic_smem_size` | LLVM `llvm.nvvm.read.ptx.sreg.dynamic_smem_size` from `int_nvvm_read_ptx_sreg_dynamic_smem_size` | safe; scope cta; pure; memory none; convergent false | PTX 4.1 on sm_20+ ([PTX ISA 9.3, 10.32 Special Registers: %dynamic_smem_size](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-dynamic-smem-size)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %dynamic_smem_size;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::elect_sync` | `nvvm.elect_sync` | LLVM `llvm.nvvm.elect.sync` from `int_nvvm_elect_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.14.15 Parallel Synchronization and Communication Instructions: elect.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-elect-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `elect.sync <register\|predicate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::grid::envreg1` | `nvvm.read_ptx_sreg_envreg1` | LLVM `llvm.nvvm.read.ptx.sreg.envreg1` from `int_nvvm_read_ptx_sreg_envreg1` | safe; scope grid; pure; memory none; convergent false | PTX 2.1 on all ([PTX ISA 9.3, 10.27 Special Registers: %envreg<32>](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-envreg)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.b32 <register>, %envreg1;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::grid::envreg2` | `nvvm.read_ptx_sreg_envreg2` | LLVM `llvm.nvvm.read.ptx.sreg.envreg2` from `int_nvvm_read_ptx_sreg_envreg2` | safe; scope grid; pure; memory none; convergent false | PTX 2.1 on all ([PTX ISA 9.3, 10.27 Special Registers: %envreg<32>](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-envreg)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.b32 <register>, %envreg2;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::ex2_approx_f16` | `nvvm.scalar_math` | LLVM `llvm.nvvm.ex2.approx` from `int_nvvm_ex2_approx` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_75+ ([PTX ISA 9.3, 9.7.4.10 Half Precision Floating Point Instructions: ex2](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-ex2)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ex2.approx.f16 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::ex2_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.ex2.approx` from `int_nvvm_ex2_approx` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.13 Floating Point Instructions: ex2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-ex2)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ex2.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::ex2_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.ex2.approx.ftz` from `int_nvvm_ex2_approx_ftz` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.13 Floating Point Instructions: ex2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-ex2)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ex2.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::fence_mbarrier_init_release_cluster` | `nvvm.fence_mbarrier_init_release_cluster` | LLVM `llvm.nvvm.fence.mbarrier_init.release.cluster` from `int_nvvm_fence_mbarrier_init_release_cluster` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.mbarrier_init](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-mbarrier-init)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fence.mbarrier_init.release.cluster;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::fence_proxy_async_generic_acquire_shared_cluster_cluster` | `nvvm.fence_proxy_async_generic_acquire_shared_cluster_cluster` | LLVM `llvm.nvvm.fence.proxy.async_generic.acquire.sync_restrict.space.cluster.scope.cluster` from `int_nvvm_fence_proxy_async_generic_acquire_sync_restrict_space_cluster_scope_cluster` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.6 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.proxy](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fence.proxy.async::generic.acquire.sync_restrict::shared::cluster.cluster;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::fence_proxy_async_generic_release_shared_cta_cluster` | `nvvm.fence_proxy_async_generic_release_shared_cta_cluster` | LLVM `llvm.nvvm.fence.proxy.async_generic.release.sync_restrict.space.cta.scope.cluster` from `int_nvvm_fence_proxy_async_generic_release_sync_restrict_space_cta_scope_cluster` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.6 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.proxy](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fence.proxy.async::generic.release.sync_restrict::shared::cta.cluster;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::fence_proxy_async_shared_cta` | `nvvm.fence_proxy_async_shared_cta` | LLVM `llvm.nvvm.fence.proxy.async.shared_cta` from `int_nvvm_fence_proxy_async_shared_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.proxy](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fence.proxy.async.shared::cta;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_cluster` | `nvvm.fence_proxy_tensormap_generic_acquire_cluster` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.cluster` from `int_nvvm_fence_proxy_tensormap_generic_acquire_cluster` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.acquire.cluster [<address>], 128;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_cta` | `nvvm.fence_proxy_tensormap_generic_acquire_cta` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.cta` from `int_nvvm_fence_proxy_tensormap_generic_acquire_cta` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.acquire.cta [<address>], 128;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_gpu` | `nvvm.fence_proxy_tensormap_generic_acquire_gpu` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.gpu` from `int_nvvm_fence_proxy_tensormap_generic_acquire_gpu` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.acquire.gpu [<address>], 128;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_system` | `nvvm.fence_proxy_tensormap_generic_acquire_system` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.sys` from `int_nvvm_fence_proxy_tensormap_generic_acquire_sys` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.acquire.sys [<address>], 128;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_cluster` | `nvvm.fence_proxy_tensormap_generic_release_cluster` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.cluster` from `int_nvvm_fence_proxy_tensormap_generic_release_cluster` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.release.cluster;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_cta` | `nvvm.fence_proxy_tensormap_generic_release_cta` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.cta` from `int_nvvm_fence_proxy_tensormap_generic_release_cta` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.release.cta;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_gpu` | `nvvm.fence_proxy_tensormap_generic_release_gpu` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.gpu` from `int_nvvm_fence_proxy_tensormap_generic_release_gpu` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.release.gpu;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_system` | `nvvm.fence_proxy_tensormap_generic_release_system` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.sys` from `int_nvvm_fence_proxy_tensormap_generic_release_sys` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `fence.proxy.tensormap::generic.release.sys;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::fma_bf16x2` | `nvvm.fma_bf16x2` | LLVM `llvm.nvvm.fma.rn.bf16x2` from `int_nvvm_fma_rn_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.bf16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::fma_f16x2` | `nvvm.fma_f16x2` | LLVM `llvm.nvvm.fma.rn.f16x2` from `int_nvvm_fma_rn_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::fma_f32x2` | `nvvm.fma_f32x2` | PTX-native `fma.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.4 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.f32x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::fma_ftz_f16x2` | `nvvm.fma_ftz_f16x2` | LLVM `llvm.nvvm.fma.rn.ftz.f16x2` from `int_nvvm_fma_rn_ftz_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.ftz.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::fma_ftz_f32x2` | `nvvm.fma_ftz_f32x2` | PTX-native `fma.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.4 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.ftz.f32x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::fma_ftz_relu_f16x2` | `nvvm.fma_ftz_relu_f16x2` | LLVM `llvm.nvvm.fma.rn.ftz.relu.f16x2` from `int_nvvm_fma_rn_ftz_relu_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.ftz.relu.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::fma_ftz_sat_f16x2` | `nvvm.fma_ftz_sat_f16x2` | LLVM `llvm.nvvm.fma.rn.ftz.sat.f16x2` from `int_nvvm_fma_rn_ftz_sat_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.ftz.sat.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::fma_relu_bf16x2` | `nvvm.fma_relu_bf16x2` | LLVM `llvm.nvvm.fma.rn.relu.bf16x2` from `int_nvvm_fma_rn_relu_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.relu.bf16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::fma_relu_f16x2` | `nvvm.fma_relu_f16x2` | LLVM `llvm.nvvm.fma.rn.relu.f16x2` from `int_nvvm_fma_rn_relu_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.relu.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.f` from `int_nvvm_fma_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rm.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.d` from `int_nvvm_fma_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rm.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.ftz.f` from `int_nvvm_fma_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rm.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rm_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.ftz.sat.f` from `int_nvvm_fma_rm_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rm.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rm_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.sat.f` from `int_nvvm_fma_rm_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rm.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.f` from `int_nvvm_fma_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.d` from `int_nvvm_fma_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.ftz.f` from `int_nvvm_fma_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rn_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.ftz.sat.f` from `int_nvvm_fma_rn_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rn_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.sat.f` from `int_nvvm_fma_rn_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.f` from `int_nvvm_fma_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rp.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.d` from `int_nvvm_fma_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rp.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.ftz.f` from `int_nvvm_fma_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rp.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rp_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.ftz.sat.f` from `int_nvvm_fma_rp_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rp.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rp_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.sat.f` from `int_nvvm_fma_rp_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rp.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.f` from `int_nvvm_fma_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.d` from `int_nvvm_fma_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rz.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.ftz.f` from `int_nvvm_fma_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rz.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rz_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.ftz.sat.f` from `int_nvvm_fma_rz_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rz.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::fma_rz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.sat.f` from `int_nvvm_fma_rz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::fma_sat_f16x2` | `nvvm.fma_sat_f16x2` | LLVM `llvm.nvvm.fma.rn.sat.f16x2` from `int_nvvm_fma_rn_sat_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `fma.rn.sat.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::debug::globaltimer` | `nvvm.read_ptx_sreg_globaltimer` | LLVM `llvm.nvvm.read.ptx.sreg.globaltimer` from `int_nvvm_read_ptx_sreg_globaltimer` | safe; scope device; impure; memory inaccessible_read_write; convergent false | PTX 3.1 on sm_30+ ([PTX ISA 9.3, 10.28 Special Registers: %globaltimer, %globaltimer_lo, %globaltimer_hi](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-globaltimer)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u64 <register>, %globaltimer;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::grid::grid_dependency_launch_dependents` | `nvvm.grid_dependency_launch_dependents` | LLVM `llvm.nvvm.griddepcontrol.launch.dependents` from `int_nvvm_griddepcontrol_launch_dependents` | unsafe; scope grid; impure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.15.1 Parallel Synchronization and Communication Instructions: griddepcontrol](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `griddepcontrol.launch_dependents;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::grid::grid_dependency_wait` | `nvvm.grid_dependency_wait` | LLVM `llvm.nvvm.griddepcontrol.wait` from `int_nvvm_griddepcontrol_wait` | unsafe; scope grid; impure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.15.1 Parallel Synchronization and Communication Instructions: griddepcontrol](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `griddepcontrol.wait;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::grid_dim_x` | `nvvm.read_ptx_sreg_nctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.nctaid.x` from `int_nvvm_read_ptx_sreg_nctaid_x` | safe; scope grid; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.7 Special Registers: %nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nctaid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::grid_dim_y` | `nvvm.read_ptx_sreg_nctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.nctaid.y` from `int_nvvm_read_ptx_sreg_nctaid_y` | safe; scope grid; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.7 Special Registers: %nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nctaid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::grid_dim_z` | `nvvm.read_ptx_sreg_nctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.nctaid.z` from `int_nvvm_read_ptx_sreg_nctaid_z` | safe; scope grid; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.7 Special Registers: %nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nctaid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::thread::gridid` | `nvvm.read_ptx_sreg_gridid` | PTX-native `mov.u64 %gridid` | safe; scope grid; pure; memory none; convergent false | PTX 3.0 on sm_30+ ([PTX ISA 9.3, 10.10 Special Registers: %gridid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-gridid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u64 <register>, %gridid;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::lane_id` | `nvvm.read_ptx_sreg_laneid` | LLVM `llvm.nvvm.read.ptx.sreg.laneid` from `int_nvvm_read_ptx_sreg_laneid` | safe; scope thread; pure; memory none; convergent false | PTX 1.3 on all ([PTX ISA 9.3, 10.3 Special Registers: %laneid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-laneid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %laneid;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::lanemask_eq` | `nvvm.read_ptx_sreg_lanemask_eq` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.eq` from `int_nvvm_read_ptx_sreg_lanemask_eq` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.11 Special Registers: %lanemask_eq](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-eq)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %lanemask_eq;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::lanemask_ge` | `nvvm.read_ptx_sreg_lanemask_ge` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.ge` from `int_nvvm_read_ptx_sreg_lanemask_ge` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.14 Special Registers: %lanemask_ge](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-ge)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %lanemask_ge;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::lanemask_gt` | `nvvm.read_ptx_sreg_lanemask_gt` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.gt` from `int_nvvm_read_ptx_sreg_lanemask_gt` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.15 Special Registers: %lanemask_gt](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-gt)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %lanemask_gt;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::lanemask_le` | `nvvm.read_ptx_sreg_lanemask_le` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.le` from `int_nvvm_read_ptx_sreg_lanemask_le` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.12 Special Registers: %lanemask_le](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-le)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %lanemask_le;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::lanemask_lt` | `nvvm.read_ptx_sreg_lanemask_lt` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.lt` from `int_nvvm_read_ptx_sreg_lanemask_lt` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.13 Special Registers: %lanemask_lt](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-lt)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %lanemask_lt;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m16n16_x1_trans_b8` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x1.trans.b8` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x1_trans_b8` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m16n16.x1.trans.shared.b8 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x1.trans.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x1_trans_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m16n16.x1.trans.shared.b8x16.b4x16_p64 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x1.trans.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x1_trans_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m16n16.x1.trans.shared.b8x16.b6x16_p32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m16n16_x2_trans_b8` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x2.trans.b8` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x2_trans_b8` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m16n16.x2.trans.shared.b8 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x2.trans.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x2_trans_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m16n16.x2.trans.shared.b8x16.b4x16_p64 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x2.trans.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x2_trans_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m16n16.x2.trans.shared.b8x16.b6x16_p32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n16_x1_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x1.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x1_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n16.x1.shared.b8x16.b4x16_p64 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n16_x1_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x1.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x1_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n16.x1.shared.b8x16.b6x16_p32 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n16_x2_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x2.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x2_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n16.x2.shared.b8x16.b4x16_p64 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n16_x2_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x2.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x2_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n16.x2.shared.b8x16.b6x16_p32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n16_x4_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x4.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x4_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n16.x4.shared.b8x16.b4x16_p64 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n16_x4_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x4.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x4_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n16.x4.shared.b8x16.b6x16_p32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n8_x1_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x1_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n8.x1.shared.b16 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n8_x1_trans_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.trans.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x1_trans_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n8.x1.trans.shared.b16 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n8_x2_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x2_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n8.x2.shared.b16 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n8_x2_trans_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x2_trans_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n8_x4_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x4_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n8.x4.shared.b16 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::ldmatrix_m8n8_x4_trans_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x4_trans_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::lg2_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.lg2.approx.f` from `int_nvvm_lg2_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.12 Floating Point Instructions: lg2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-lg2)) | validated on `rust-llvm-23.1.0-16696adc`; expects `lg2.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::lg2_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.lg2.approx.ftz.f` from `int_nvvm_lg2_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.12 Floating Point Instructions: lg2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-lg2)) | validated on `rust-llvm-23.1.0-16696adc`; expects `lg2.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::cluster::map_shared_rank` | `nvvm.mapa_shared_cluster` | LLVM `llvm.nvvm.mapa.shared.cluster` from `int_nvvm_mapa_shared_cluster` | unsafe; scope cluster; impure; memory none; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.24 Data Movement and Conversion Instructions: mapa](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-mapa)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mapa.shared::cluster.u64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::match_all_i64_sync` | `nvvm.match_all_sync_i64` | LLVM `llvm.nvvm.match.all.sync.i64p` from `int_nvvm_match_all_sync_i64p` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `match.all.sync.b64 <register\|predicate>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::match_all_sync` | `nvvm.match_all_sync_i32` | LLVM `llvm.nvvm.match.all.sync.i32p` from `int_nvvm_match_all_sync_i32p` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `match.all.sync.b32 <register\|predicate>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::match_any_i64_sync` | `nvvm.match_any_sync_i64` | LLVM `llvm.nvvm.match.any.sync.i64` from `int_nvvm_match_any_sync_i64` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `match.any.sync.b64 <register>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::match_any_sync` | `nvvm.match_any_sync_i32` | LLVM `llvm.nvvm.match.any.sync.i32` from `int_nvvm_match_any_sync_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `match.any.sync.b32 <register>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::max_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.bf16` from `int_nvvm_fmax_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::max_bf16x2` | `nvvm.max_bf16x2` | LLVM `llvm.nvvm.fmax.bf16x2` from `int_nvvm_fmax_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.f16` from `int_nvvm_fmax_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_f16x2` | `nvvm.max_f16x2` | LLVM `llvm.nvvm.fmax.f16x2` from `int_nvvm_fmax_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_ftz_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.f16` from `int_nvvm_fmax_ftz_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_ftz_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.f16x2` from `int_nvvm_fmax_ftz_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_ftz_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.f16` from `int_nvvm_fmax_ftz_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_ftz_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.f16x2` from `int_nvvm_fmax_ftz_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_ftz_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16` from `int_nvvm_fmax_ftz_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_ftz_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16x2` from `int_nvvm_fmax_ftz_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::max_ftz_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.xorsign.abs.f` from `int_nvvm_fmax_ftz_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_ftz_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.xorsign.abs.f16` from `int_nvvm_fmax_ftz_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_ftz_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.xorsign.abs.f16x2` from `int_nvvm_fmax_ftz_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::max_ftz_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.xorsign.abs.f` from `int_nvvm_fmax_ftz_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.ftz.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::max_nan_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.bf16` from `int_nvvm_fmax_nan_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::max_nan_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.bf16x2` from `int_nvvm_fmax_nan_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.f16` from `int_nvvm_fmax_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.f16x2` from `int_nvvm_fmax_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::max_nan_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.bf16` from `int_nvvm_fmax_nan_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::max_nan_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.bf16x2` from `int_nvvm_fmax_nan_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.f16` from `int_nvvm_fmax_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.f16x2` from `int_nvvm_fmax_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::max_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.f` from `int_nvvm_fmax_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::i16x2::max_relu_s16x2` | `nvvm.max_relu_s16x2` | PTX-native `max.relu.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `max.relu.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::int::max_relu_s32` | `nvvm.max_relu_s32` | PTX-native `max.relu.s32` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `max.relu.s32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::i16x2::max_s16x2` | `nvvm.max_s16x2` | PTX-native `max.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `max.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::i16x2::max_u16x2` | `nvvm.max_u16x2` | PTX-native `max.u16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `max.u16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::max_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.bf16` from `int_nvvm_fmax_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::max_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.bf16x2` from `int_nvvm_fmax_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::max_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.f16` from `int_nvvm_fmax_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::max_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.f16x2` from `int_nvvm_fmax_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::max_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.f` from `int_nvvm_fmax_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.0-16696adc`; expects `max.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_arrive` | `nvvm.mbarrier_arrive_shared` | LLVM `llvm.nvvm.mbarrier.arrive.shared` from `int_nvvm_mbarrier_arrive_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.16 Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.arrive.shared.b64 <register>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_arrive_cluster` | `nvvm.mbarrier_arrive_cluster` | PTX-native `mbarrier.arrive.release.cluster.shared::cluster.b64` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.arrive.release.cluster.shared::cluster.b64 _, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_arrive_expect_tx` | `nvvm.mbarrier_arrive_expect_tx_shared` | LLVM `llvm.nvvm.mbarrier.arrive.expect.tx.scope.cta.space.cta` from `int_nvvm_mbarrier_arrive_expect_tx_scope_cta_space_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.arrive.expect_tx.release.cta.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_arrive_expect_tx_cluster` | `nvvm.mbarrier_arrive_expect_tx_cluster` | LLVM `llvm.nvvm.mbarrier.arrive.expect.tx.relaxed.scope.cluster.space.cta` from `int_nvvm_mbarrier_arrive_expect_tx_relaxed_scope_cluster_space_cta` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.6 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.arrive.expect_tx.relaxed.cluster.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_arrive_no_complete` | `nvvm.mbarrier_arrive_no_complete_shared` | LLVM `llvm.nvvm.mbarrier.arrive.noComplete.shared` from `int_nvvm_mbarrier_arrive_noComplete_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.16 Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.arrive.noComplete.shared.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_init` | `nvvm.mbarrier_init_shared` | LLVM `llvm.nvvm.mbarrier.init.shared` from `int_nvvm_mbarrier_init_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.12 Parallel Synchronization and Communication Instructions: mbarrier.init](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-init)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.init.shared.b64 [<address>], <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_inval` | `nvvm.mbarrier_inval_shared` | LLVM `llvm.nvvm.mbarrier.inval.shared` from `int_nvvm_mbarrier_inval_shared` | unsafe; scope cta; impure; memory write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.13 Parallel Synchronization and Communication Instructions: mbarrier.inval](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-inval)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.inval.shared.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_test_wait` | `nvvm.mbarrier_test_wait_shared` | LLVM `llvm.nvvm.mbarrier.test.wait.shared` from `int_nvvm_mbarrier_test_wait_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.19 Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.test_wait.shared.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_try_wait` | `nvvm.mbarrier_try_wait_shared` | LLVM `llvm.nvvm.mbarrier.try.wait.scope.cta.space.cta` from `int_nvvm_mbarrier_try_wait_scope_cta_space_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.try_wait.shared.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_try_wait_parity` | `nvvm.mbarrier_try_wait_parity_shared` | LLVM `llvm.nvvm.mbarrier.try.wait.parity.scope.cta.space.cta` from `int_nvvm_mbarrier_try_wait_parity_scope_cta_space_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.try_wait.parity.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::mbarrier_try_wait_parity_cluster` | `nvvm.mbarrier_try_wait_parity_cluster` | LLVM `llvm.nvvm.mbarrier.try.wait.parity.scope.cluster.space.cta` from `int_nvvm_mbarrier_try_wait_parity_scope_cluster_space_cta` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mbarrier.try_wait.parity.acquire.cluster.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::min_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.bf16` from `int_nvvm_fmin_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::min_bf16x2` | `nvvm.min_bf16x2` | LLVM `llvm.nvvm.fmin.bf16x2` from `int_nvvm_fmin_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.f16` from `int_nvvm_fmin_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_f16x2` | `nvvm.min_f16x2` | LLVM `llvm.nvvm.fmin.f16x2` from `int_nvvm_fmin_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_ftz_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.f16` from `int_nvvm_fmin_ftz_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_ftz_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.f16x2` from `int_nvvm_fmin_ftz_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_ftz_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.f16` from `int_nvvm_fmin_ftz_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_ftz_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.f16x2` from `int_nvvm_fmin_ftz_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_ftz_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16` from `int_nvvm_fmin_ftz_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_ftz_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16x2` from `int_nvvm_fmin_ftz_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::min_ftz_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.xorsign.abs.f` from `int_nvvm_fmin_ftz_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_ftz_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.xorsign.abs.f16` from `int_nvvm_fmin_ftz_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_ftz_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.xorsign.abs.f16x2` from `int_nvvm_fmin_ftz_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::min_ftz_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.xorsign.abs.f` from `int_nvvm_fmin_ftz_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.ftz.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::min_nan_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.bf16` from `int_nvvm_fmin_nan_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::min_nan_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.bf16x2` from `int_nvvm_fmin_nan_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.f16` from `int_nvvm_fmin_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.f16x2` from `int_nvvm_fmin_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::min_nan_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.bf16` from `int_nvvm_fmin_nan_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::min_nan_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.bf16x2` from `int_nvvm_fmin_nan_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.f16` from `int_nvvm_fmin_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.f16x2` from `int_nvvm_fmin_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::min_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.f` from `int_nvvm_fmin_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::i16x2::min_relu_s16x2` | `nvvm.min_relu_s16x2` | PTX-native `min.relu.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `min.relu.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::int::min_relu_s32` | `nvvm.min_relu_s32` | PTX-native `min.relu.s32` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `min.relu.s32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::i16x2::min_s16x2` | `nvvm.min_s16x2` | PTX-native `min.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `min.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::i16x2::min_u16x2` | `nvvm.min_u16x2` | PTX-native `min.u16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `min.u16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16::min_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.bf16` from `int_nvvm_fmin_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::min_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.bf16x2` from `int_nvvm_fmin_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16::min_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.f16` from `int_nvvm_fmin_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::min_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.f16x2` from `int_nvvm_fmin_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::min_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.f` from `int_nvvm_fmin_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.0-16696adc`; expects `min.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k128_s32_b1_and_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.and.popc.m16n8k128.row.col.b1` from `int_nvvm_mma_and_popc_m16n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k128.row.col.s32.b1.b1.s32.and.popc <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k128_s32_b1_xor_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.xor.popc.m16n8k128.row.col.b1` from `int_nvvm_mma_xor_popc_m16n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k128.row.col.s32.b1.b1.s32.xor.popc <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_f16_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_f32_bf16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.bf16` from `int_nvvm_mma_m16n8k16_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_f32_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e4m3.e4m3.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e4m3.e5m2.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e5m2.e4m3.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e5m2.e5m2.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e4m3.e4m3.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e4m3.e5m2.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e5m2.e4m3.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e5m2.e5m2.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.s8` from `int_nvvm_mma_m16n8k16_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.s8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.s8.u8` from `int_nvvm_mma_m16n8k16_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.s8.u8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.u8` from `int_nvvm_mma_m16n8k16_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.u8.s8` from `int_nvvm_mma_m16n8k16_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.u8.s8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.u8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k256_s32_b1_and_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.and.popc.m16n8k256.row.col.b1` from `int_nvvm_mma_and_popc_m16n8k256_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k256.row.col.s32.b1.b1.s32.and.popc <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k256_s32_b1_xor_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.xor.popc.m16n8k256.row.col.b1` from `int_nvvm_mma_xor_popc_m16n8k256_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k256.row.col.s32.b1.b1.s32.xor.popc <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e4m3.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e4m3.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e5m2.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e5m2.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e4m3.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e4m3.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e5m2.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e5m2.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s4` from `int_nvvm_mma_m16n8k32_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s4.u4` from `int_nvvm_mma_m16n8k32_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s4.u4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s8` from `int_nvvm_mma_m16n8k32_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s8.u8` from `int_nvvm_mma_m16n8k32_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s8.u8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u4` from `int_nvvm_mma_m16n8k32_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u4.s4` from `int_nvvm_mma_m16n8k32_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u4.s4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u8` from `int_nvvm_mma_m16n8k32_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u8.s8` from `int_nvvm_mma_m16n8k32_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u8.s8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k4_f32_tf32` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k4.row.col.tf32` from `int_nvvm_mma_m16n8k4_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k4.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.s4` from `int_nvvm_mma_m16n8k64_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.s4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.s4.u4` from `int_nvvm_mma_m16n8k64_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.s4.u4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.u4` from `int_nvvm_mma_m16n8k64_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.u4.s4` from `int_nvvm_mma_m16n8k64_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.u4.s4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.u4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k8_f16_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.f16.f16` from `int_nvvm_mma_m16n8k8_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k8.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k8_f32_bf16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.bf16` from `int_nvvm_mma_m16n8k8_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k8.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k8_f32_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.f32.f32` from `int_nvvm_mma_m16n8k8_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m16n8k8_f32_tf32` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.tf32` from `int_nvvm_mma_m16n8k8_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k128_s32_b1_and_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.and.popc.m8n8k128.row.col.b1` from `int_nvvm_mma_and_popc_m8n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k128.row.col.s32.b1.b1.s32.and.popc <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k128_s32_b1_xor_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.xor.popc.m8n8k128.row.col.b1` from `int_nvvm_mma_xor_popc_m8n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k128.row.col.s32.b1.b1.s32.xor.popc <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.s8` from `int_nvvm_mma_m8n8k16_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.s32.s8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.s8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.s8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.s8.u8` from `int_nvvm_mma_m8n8k16_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.s32.s8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.s8.u8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.s8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.u8` from `int_nvvm_mma_m8n8k16_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.s32.u8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.u8.s8` from `int_nvvm_mma_m8n8k16_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.s32.u8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.u8.s8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.u8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.u8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.u8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.s4` from `int_nvvm_mma_m8n8k32_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.s32.s4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.s4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.s4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.s4.u4` from `int_nvvm_mma_m8n8k32_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.s32.s4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.s4.u4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.s4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.u4` from `int_nvvm_mma_m8n8k32_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.s32.u4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.u4.s4` from `int_nvvm_mma_m8n8k32_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.s32.u4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.u4.s4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.u4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.u4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.u4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_m8n8k4_f64` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k4.row.col.f64` from `int_nvvm_mma_m8n8k4_row_col_f64` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sync.aligned.m8n8k4.row.col.f64.f64.f64.f64 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.s4.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.u4.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.s8.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.u8.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e4m3_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e4m3.e4m3.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e4m3_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e4m3.e5m2.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e5m2_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e5m2.e4m3.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e5m2_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e5m2.e5m2.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s4.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s8.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u4.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u8.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f16_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.f16.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:2>, <register-list:2>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f32_bf16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.bf16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f32_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.f32.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f32_tf32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.tf32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_f16_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.f16.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_f32_bf16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.bf16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_f32_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.f32.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k8_f32_tf32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k8.row.col.tf32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k8_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | executed on `rust-llvm-23.1.0-16696adc`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::movmatrix_trans_b16` | `nvvm.movmatrix_trans_b16` | PTX-native `movmatrix.sync.aligned.m8n8.trans.b16` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.8 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.17 Warp-level matrix transpose instruction: movmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-movmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `movmatrix.sync.aligned.m8n8.trans.b16 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::mul_bf16x2` | `nvvm.mul_bf16x2` | PTX-native `mul.rn.bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.4.3 Half Precision Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::mul_f16x2` | `nvvm.mul_f16x2` | PTX-native `mul.rn.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.3 Half Precision Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::mul_f32x2` | `nvvm.mul_f32x2` | PTX-native `mul.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::mul_ftz_f32x2` | `nvvm.mul_ftz_f32x2` | PTX-native `mul.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.ftz.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rm.f` from `int_nvvm_mul_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rm.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rm.d` from `int_nvvm_mul_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rm.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rm.ftz.f` from `int_nvvm_mul_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rm.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rn.f` from `int_nvvm_mul_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rn.d` from `int_nvvm_mul_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rn.ftz.f` from `int_nvvm_mul_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rn.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rp.f` from `int_nvvm_mul_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rp.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rp.d` from `int_nvvm_mul_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rp.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rp.ftz.f` from `int_nvvm_mul_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rp.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rz.f` from `int_nvvm_mul_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rz.d` from `int_nvvm_mul_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rz.f64 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::mul_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rz.ftz.f` from `int_nvvm_mul_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mul.rz.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::barrier::nanosleep` | `nvvm.nanosleep` | LLVM `llvm.nvvm.nanosleep` from `int_nvvm_nanosleep` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 6.3 on sm_70+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: nanosleep](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-nanosleep)) | validated on `rust-llvm-23.1.0-16696adc`; expects `nanosleep.u32 <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::neg_bf16x2` | `nvvm.neg_bf16x2` | LLVM `llvm.nvvm.neg.bf16x2` from `int_nvvm_neg_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.5 Half Precision Floating Point Instructions: neg](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg)) | validated on `rust-llvm-23.1.0-16696adc`; expects `neg.bf16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::neg_f16x2` | `nvvm.neg_f16x2` | PTX-native `neg.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.4.5 Half Precision Floating Point Instructions: neg](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg)) | validated on `rust-llvm-23.1.0-16696adc`; expects `neg.f16x2 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::thread::nsmid` | `nvvm.read_ptx_sreg_nsmid` | LLVM `llvm.nvvm.read.ptx.sreg.nsmid` from `int_nvvm_read_ptx_sreg_nsmid` | safe; scope device; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.9 Special Registers: %nsmid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nsmid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nsmid;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::nwarpid` | `nvvm.read_ptx_sreg_nwarpid` | LLVM `llvm.nvvm.read.ptx.sreg.nwarpid` from `int_nvvm_read_ptx_sreg_nwarpid` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.5 Special Registers: %nwarpid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nwarpid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %nwarpid;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::atomic::atom_add_bf16x2` | `nvvm.packed_atomic_add` | PTX-native `atom.global.add.noftz.bf16x2` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.13.1.1 Atomic reduction and atomic operations: atom](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom)) | validated on `rust-llvm-23.1.0-16696adc`; expects `atom.global.add.noftz.bf16x2 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::atomic::atom_add_f16x2` | `nvvm.packed_atomic_add` | PTX-native `atom.global.add.noftz.f16x2` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 6.2 on sm_70+ ([PTX ISA 9.3, 9.7.13.1.1 Atomic reduction and atomic operations: atom](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom)) | validated on `rust-llvm-23.1.0-16696adc`; expects `atom.global.add.noftz.f16x2 <register>, [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::debug::pmevent` | `nvvm.pmevent` | PTX-native `pmevent` | safe; scope thread; impure; memory none; convergent false | PTX 1.4 on all ([PTX ISA 9.3, 9.7.20.3 Miscellaneous Instructions: pmevent](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-pmevent)) | validated on `rust-llvm-23.1.0-16696adc`; expects `pmevent <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::prefetch_tma_descriptor` | `nvvm.prefetch_tensormap` | LLVM `llvm.nvvm.prefetch.tensormap` from `int_nvvm_prefetch_tensormap` | unsafe; scope thread; impure; memory read; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `prefetch.tensormap [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt` from `int_nvvm_prmt` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt_b4e` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.b4e` from `int_nvvm_prmt_b4e` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32.b4e <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt_ecl` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.ecl` from `int_nvvm_prmt_ecl` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32.ecl <register>, <register>, 0, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt_ecr` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.ecr` from `int_nvvm_prmt_ecr` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32.ecr <register>, <register>, 0, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt_f4e` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.f4e` from `int_nvvm_prmt_f4e` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32.f4e <register>, <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt_rc16` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.rc16` from `int_nvvm_prmt_rc16` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32.rc16 <register>, <register>, 0, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::prmt::prmt_rc8` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.rc8` from `int_nvvm_prmt_rc8` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `prmt.b32.rc8 <register>, <register>, 0, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.approx.ftz.f` from `int_nvvm_rcp_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_approx_ftz_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.approx.ftz.d` from `int_nvvm_rcp_approx_ftz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.approx.ftz.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rm_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rm.f` from `int_nvvm_rcp_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rm.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rm_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rm.d` from `int_nvvm_rcp_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rm.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rm_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rm.ftz.f` from `int_nvvm_rcp_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rm.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rn_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rn.f` from `int_nvvm_rcp_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rn.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rn_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rn.d` from `int_nvvm_rcp_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rn.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rn_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rn.ftz.f` from `int_nvvm_rcp_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rn.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rp_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rp.f` from `int_nvvm_rcp_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rp.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rp_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rp.d` from `int_nvvm_rcp_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rp.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rp_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rp.ftz.f` from `int_nvvm_rcp_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rp.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rz.f` from `int_nvvm_rcp_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rz_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rz.d` from `int_nvvm_rcp_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rz.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rcp_rz_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rz.ftz.f` from `int_nvvm_rcp_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rcp.rz.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_add` | `nvvm.redux_sync_add` | LLVM `llvm.nvvm.redux.sync.add` from `int_nvvm_redux_sync_add` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.add.s32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_and` | `nvvm.redux_sync_and` | LLVM `llvm.nvvm.redux.sync.and` from `int_nvvm_redux_sync_and` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.and.b32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_max_abs_f32` | `nvvm.redux_sync_fmax_abs` | LLVM `llvm.nvvm.redux.sync.fmax.abs` from `int_nvvm_redux_sync_fmax_abs` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.max.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_max_abs_nan_f32` | `nvvm.redux_sync_fmax_abs_nan` | LLVM `llvm.nvvm.redux.sync.fmax.abs.NaN` from `int_nvvm_redux_sync_fmax_abs_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.max.abs.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_max_f32` | `nvvm.redux_sync_fmax` | LLVM `llvm.nvvm.redux.sync.fmax` from `int_nvvm_redux_sync_fmax` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.max.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_max_i32` | `nvvm.redux_sync_max` | LLVM `llvm.nvvm.redux.sync.max` from `int_nvvm_redux_sync_max` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.max.s32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_max_nan_f32` | `nvvm.redux_sync_fmax_nan` | LLVM `llvm.nvvm.redux.sync.fmax.NaN` from `int_nvvm_redux_sync_fmax_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.max.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_max_u32` | `nvvm.redux_sync_umax` | LLVM `llvm.nvvm.redux.sync.umax` from `int_nvvm_redux_sync_umax` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.max.u32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_min_abs_f32` | `nvvm.redux_sync_fmin_abs` | LLVM `llvm.nvvm.redux.sync.fmin.abs` from `int_nvvm_redux_sync_fmin_abs` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.min.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_min_abs_nan_f32` | `nvvm.redux_sync_fmin_abs_nan` | LLVM `llvm.nvvm.redux.sync.fmin.abs.NaN` from `int_nvvm_redux_sync_fmin_abs_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.min.abs.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_min_f32` | `nvvm.redux_sync_fmin` | LLVM `llvm.nvvm.redux.sync.fmin` from `int_nvvm_redux_sync_fmin` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.min.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_min_i32` | `nvvm.redux_sync_min` | LLVM `llvm.nvvm.redux.sync.min` from `int_nvvm_redux_sync_min` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.min.s32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_min_nan_f32` | `nvvm.redux_sync_fmin_nan` | LLVM `llvm.nvvm.redux.sync.fmin.NaN` from `int_nvvm_redux_sync_fmin_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.min.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_min_u32` | `nvvm.redux_sync_umin` | LLVM `llvm.nvvm.redux.sync.umin` from `int_nvvm_redux_sync_umin` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.min.u32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_or` | `nvvm.redux_sync_or` | LLVM `llvm.nvvm.redux.sync.or` from `int_nvvm_redux_sync_or` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.or.b32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::redux_sync_xor` | `nvvm.redux_sync_xor` | LLVM `llvm.nvvm.redux.sync.xor` from `int_nvvm_redux_sync_xor` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `redux.sync.xor.b32 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rsqrt_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rsqrt.approx.f` from `int_nvvm_rsqrt_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.14 Floating Point Instructions: rsqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rsqrt.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rsqrt_approx_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rsqrt.approx.d` from `int_nvvm_rsqrt_approx_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.14 Floating Point Instructions: rsqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rsqrt.approx.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::rsqrt_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rsqrt.approx.ftz.f` from `int_nvvm_rsqrt_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.14 Floating Point Instructions: rsqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `rsqrt.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::thread::setmaxnreg_dec` | `nvvm.setmaxnreg_dec` | LLVM `llvm.nvvm.setmaxnreg.dec.sync.aligned.u32` from `int_nvvm_setmaxnreg_dec_sync_aligned_u32` | unsafe; scope warpgroup; impure; memory none; convergent true | PTX 8.0 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, 9.7.16.5 Miscellaneous Instructions: setmaxnreg](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `setmaxnreg.dec.sync.aligned.u32 <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::thread::setmaxnreg_inc` | `nvvm.setmaxnreg_inc` | LLVM `llvm.nvvm.setmaxnreg.inc.sync.aligned.u32` from `int_nvvm_setmaxnreg_inc_sync_aligned_u32` | unsafe; scope warpgroup; impure; memory none; convergent true | PTX 8.0 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, 9.7.16.5 Miscellaneous Instructions: setmaxnreg](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg)) | validated on `rust-llvm-23.1.0-16696adc-execution-control`; expects `setmaxnreg.inc.sync.aligned.u32 <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_down_f32_sync` | `nvvm.shfl_sync_down_f32` | LLVM `llvm.nvvm.shfl.sync.down.f32` from `int_nvvm_shfl_sync_down_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.down.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_down_sync` | `nvvm.shfl_sync_down_i32` | LLVM `llvm.nvvm.shfl.sync.down.i32` from `int_nvvm_shfl_sync_down_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.down.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_down_u64_sync` | `nvvm.shfl_sync_down_i64` | PTX-native `shfl.sync.down.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.down.b32 lo, lo, <register>, 31, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_f32_sync` | `nvvm.shfl_sync_idx_f32` | LLVM `llvm.nvvm.shfl.sync.idx.f32` from `int_nvvm_shfl_sync_idx_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.idx.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_sync` | `nvvm.shfl_sync_idx_i32` | LLVM `llvm.nvvm.shfl.sync.idx.i32` from `int_nvvm_shfl_sync_idx_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.idx.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_u64_sync` | `nvvm.shfl_sync_idx_i64` | PTX-native `shfl.sync.idx.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.idx.b32 lo, lo, <register>, 31, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_up_f32_sync` | `nvvm.shfl_sync_up_f32` | LLVM `llvm.nvvm.shfl.sync.up.f32` from `int_nvvm_shfl_sync_up_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.up.b32 <register>, <register>, <register-or-immediate>, 0, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_up_sync` | `nvvm.shfl_sync_up_i32` | LLVM `llvm.nvvm.shfl.sync.up.i32` from `int_nvvm_shfl_sync_up_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.up.b32 <register>, <register>, <register-or-immediate>, 0, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_up_u64_sync` | `nvvm.shfl_sync_up_i64` | PTX-native `shfl.sync.up.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.up.b32 lo, lo, <register>, 0, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_xor_f32_sync` | `nvvm.shfl_sync_bfly_f32` | LLVM `llvm.nvvm.shfl.sync.bfly.f32` from `int_nvvm_shfl_sync_bfly_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.bfly.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_xor_sync` | `nvvm.shfl_sync_bfly_i32` | LLVM `llvm.nvvm.shfl.sync.bfly.i32` from `int_nvvm_shfl_sync_bfly_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.bfly.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::shuffle_xor_u64_sync` | `nvvm.shfl_sync_bfly_i64` | PTX-native `shfl.sync.bfly.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `shfl.sync.bfly.b32 lo, lo, <register>, 31, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sin_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sin.approx.f` from `int_nvvm_sin_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: sin](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sin)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sin.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sin_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sin.approx.ftz.f` from `int_nvvm_sin_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: sin](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sin)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sin.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::thread::smid` | `nvvm.read_ptx_sreg_smid` | LLVM `llvm.nvvm.read.ptx.sreg.smid` from `int_nvvm_read_ptx_sreg_smid` | safe; scope thread; impure; memory none; convergent false | PTX 1.3 on all ([PTX ISA 9.3, 10.8 Special Registers: %smid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-smid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %smid;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.approx.f` from `int_nvvm_sqrt_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.approx.ftz.f` from `int_nvvm_sqrt_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rm_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rm.f` from `int_nvvm_sqrt_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rm.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rm_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rm.d` from `int_nvvm_sqrt_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rm.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rm_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rm.ftz.f` from `int_nvvm_sqrt_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rm.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rn_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rn.f` from `int_nvvm_sqrt_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rn.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rn_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rn.d` from `int_nvvm_sqrt_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rn.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rn_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rn.ftz.f` from `int_nvvm_sqrt_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rn.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rp_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rp.f` from `int_nvvm_sqrt_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rp.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rp_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rp.d` from `int_nvvm_sqrt_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rp.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rp_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rp.ftz.f` from `int_nvvm_sqrt_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rp.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rz.f` from `int_nvvm_sqrt_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rz_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rz.d` from `int_nvvm_sqrt_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rz.f64 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::sqrt_rz_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rz.ftz.f` from `int_nvvm_sqrt_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sqrt.rz.ftz.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::stmatrix_m8n8_x2_b16` | `nvvm.stmatrix_m8n8_x2` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x2_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `stmatrix.sync.aligned.m8n8.x2.shared.b16 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::stmatrix_m8n8_x2_trans_b16` | `nvvm.stmatrix_m8n8_x2_trans` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.trans.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x2_trans_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `stmatrix.sync.aligned.m8n8.x2.trans.shared.b16 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::stmatrix_m8n8_x4_b16` | `nvvm.stmatrix_m8n8_x4` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x4_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `stmatrix.sync.aligned.m8n8.x4.shared.b16 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::matrix::stmatrix_m8n8_x4_trans_b16` | `nvvm.stmatrix_m8n8_x4_trans` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x4_trans_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.0-16696adc`; expects `stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::bf16x2::sub_bf16x2` | `nvvm.sub_bf16x2` | PTX-native `sub.rn.bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.4.2 Half Precision Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sub.rn.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f16x2::sub_f16x2` | `nvvm.sub_f16x2` | PTX-native `sub.rn.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.2 Half Precision Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sub.rn.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::sub_f32x2` | `nvvm.sub_f32x2` | PTX-native `sub.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.2 Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sub.rn.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::f32x2::sub_ftz_f32x2` | `nvvm.sub_ftz_f32x2` | PTX-native `sub.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.2 Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub)) | validated on `rust-llvm-23.1.0-16696adc`; expects `sub.rn.ftz.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::sync_mask` | `nvvm.bar_warp_sync` | LLVM `llvm.nvvm.bar.warp.sync` from `int_nvvm_bar_warp_sync` | unsafe; scope warp; impure; memory read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.2 Parallel Synchronization and Communication Instructions: bar.warp.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-warp-sync)) | validated on `rust-llvm-23.1.0-16696adc`; expects `bar.warp.sync <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::thread::sync_threads` | `nvvm.barrier0` | LLVM `llvm.nvvm.barrier.cta.sync.aligned.all` from `int_nvvm_barrier_cta_sync_aligned_all` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 1.0 on all ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.0-16696adc`; expects `bar.sync 0;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::float::tanh_approx_f32` | `nvvm.scalar_math` | PTX-native `tanh.approx.f32` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.15 Floating Point Instructions: tanh](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-tanh)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tanh.approx.f32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_alloc` | `nvvm.tcgen05_alloc` | LLVM `llvm.nvvm.tcgen05.alloc.shared.cg1` from `int_nvvm_tcgen05_alloc_shared_cg1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_alloc_cg2` | `nvvm.tcgen05_alloc_cg2` | LLVM `llvm.nvvm.tcgen05.alloc.shared.cg2` from `int_nvvm_tcgen05_alloc_shared_cg2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.alloc.cta_group::2.sync.aligned.shared::cta.b32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_commit` | `nvvm.tcgen05_commit` | LLVM `llvm.nvvm.tcgen05.commit.cg1` from `int_nvvm_tcgen05_commit_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.commit.cta_group::1.mbarrier::arrive::one.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_commit_cg2` | `nvvm.tcgen05_commit_cg2` | LLVM `llvm.nvvm.tcgen05.commit.cg2` from `int_nvvm_tcgen05_commit_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.commit.cta_group::2.mbarrier::arrive::one.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_commit_multicast` | `nvvm.tcgen05_commit_multicast` | LLVM `llvm.nvvm.tcgen05.commit.mc.shared.cg1` from `int_nvvm_tcgen05_commit_mc_shared_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_commit_multicast_cg2` | `nvvm.tcgen05_commit_multicast_cg2` | LLVM `llvm.nvvm.tcgen05.commit.mc.shared.cg2` from `int_nvvm_tcgen05_commit_mc_shared_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_commit_shared_cluster` | `nvvm.tcgen05_commit_shared_cluster` | LLVM `llvm.nvvm.tcgen05.commit.shared.cg1` from `int_nvvm_tcgen05_commit_shared_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_commit_shared_cluster_cg2` | `nvvm.tcgen05_commit_shared_cluster_cg2` | LLVM `llvm.nvvm.tcgen05.commit.shared.cg2` from `int_nvvm_tcgen05_commit_shared_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.b64 [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b` | `nvvm.tcgen05_cp_128x128b` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.cg1` from `int_nvvm_tcgen05_cp_128x128b_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.128x128b [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b4x16_p64` | `nvvm.tcgen05_cp_128x128b_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_128x128b_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.128x128b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b4x16_p64_cg2` | `nvvm.tcgen05_cp_128x128b_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_128x128b_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.128x128b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b6x16_p32` | `nvvm.tcgen05_cp_128x128b_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_128x128b_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.128x128b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b6x16_p32_cg2` | `nvvm.tcgen05_cp_128x128b_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_128x128b_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.128x128b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_cg2` | `nvvm.tcgen05_cp_128x128b_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.cg2` from `int_nvvm_tcgen05_cp_128x128b_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.128x128b [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b4x16_p64` | `nvvm.tcgen05_cp_128x256b_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_128x256b_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.128x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b4x16_p64_cg2` | `nvvm.tcgen05_cp_128x256b_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_128x256b_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.128x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b6x16_p32` | `nvvm.tcgen05_cp_128x256b_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_128x256b_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.128x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b6x16_p32_cg2` | `nvvm.tcgen05_cp_128x256b_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_128x256b_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.128x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4` | `nvvm.tcgen05_cp_32x128b_warpx4` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.cg1` from `int_nvvm_tcgen05_cp_32x128b_warpx4_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.32x128b.warpx4 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b4x16_p64` | `nvvm.tcgen05_cp_32x128b_warpx4_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.32x128b.warpx4.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2` | `nvvm.tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.32x128b.warpx4.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b6x16_p32` | `nvvm.tcgen05_cp_32x128b_warpx4_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.32x128b.warpx4.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2` | `nvvm.tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.32x128b.warpx4.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_cg2` | `nvvm.tcgen05_cp_32x128b_warpx4_cg2` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.cg2` from `int_nvvm_tcgen05_cp_32x128b_warpx4_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.32x128b.warpx4 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b` | `nvvm.tcgen05_cp_4x256b` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.cg1` from `int_nvvm_tcgen05_cp_4x256b_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.4x256b [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b4x16_p64` | `nvvm.tcgen05_cp_4x256b_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_4x256b_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.4x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b4x16_p64_cg2` | `nvvm.tcgen05_cp_4x256b_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_4x256b_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.4x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b6x16_p32` | `nvvm.tcgen05_cp_4x256b_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_4x256b_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.4x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b6x16_p32_cg2` | `nvvm.tcgen05_cp_4x256b_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_4x256b_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.4x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_cg2` | `nvvm.tcgen05_cp_4x256b_cg2` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.cg2` from `int_nvvm_tcgen05_cp_4x256b_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.4x256b [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::01_23 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::01_23.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::01_23.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::01_23.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::01_23.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::01_23 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::02_13 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::02_13.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::02_13.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::02_13.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::02_13.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::02_13 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_smem_to_tmem` | `nvvm.tcgen05_cp_smem_to_tmem` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.cg1` from `int_nvvm_tcgen05_cp_128x256b_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::1.128x256b [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_cp_smem_to_tmem_cg2` | `nvvm.tcgen05_cp_smem_to_tmem_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.cg2` from `int_nvvm_tcgen05_cp_128x256b_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.cp.cta_group::2.128x256b [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_dealloc` | `nvvm.tcgen05_dealloc` | LLVM `llvm.nvvm.tcgen05.dealloc.cg1` from `int_nvvm_tcgen05_dealloc_cg1` | unsafe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.dealloc.cta_group::1.sync.aligned.b32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_dealloc_cg2` | `nvvm.tcgen05_dealloc_cg2` | LLVM `llvm.nvvm.tcgen05.dealloc.cg2` from `int_nvvm_tcgen05_dealloc_cg2` | unsafe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.dealloc.cta_group::2.sync.aligned.b32 <register>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_fence_after_thread_sync` | `nvvm.tcgen05_fence_after_thread_sync` | LLVM `llvm.nvvm.tcgen05.fence.after.thread.sync` from `int_nvvm_tcgen05_fence_after_thread_sync` | safe; scope thread; impure; memory none; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.fence::after_thread_sync;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_fence_before_thread_sync` | `nvvm.tcgen05_fence_before_thread_sync` | LLVM `llvm.nvvm.tcgen05.fence.before.thread.sync` from `int_nvvm_tcgen05_fence_before_thread_sync` | safe; scope thread; impure; memory none; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.fence::before_thread_sync;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x16_pack16` | `nvvm.tcgen05_ld_16x128b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x16` from `int_nvvm_tcgen05_ld_16x128b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x16.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x16_raw` | `nvvm.tcgen05_ld_16x128b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x16` from `int_nvvm_tcgen05_ld_16x128b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x16.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x1_pack16` | `nvvm.tcgen05_ld_16x128b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x1` from `int_nvvm_tcgen05_ld_16x128b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x1.pack::16b.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x1_raw` | `nvvm.tcgen05_ld_16x128b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x1` from `int_nvvm_tcgen05_ld_16x128b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x1.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x2_pack16` | `nvvm.tcgen05_ld_16x128b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x2` from `int_nvvm_tcgen05_ld_16x128b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x2.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x2_raw` | `nvvm.tcgen05_ld_16x128b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x2` from `int_nvvm_tcgen05_ld_16x128b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x2.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x32_pack16` | `nvvm.tcgen05_ld_16x128b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x32` from `int_nvvm_tcgen05_ld_16x128b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x32.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x32_raw` | `nvvm.tcgen05_ld_16x128b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x32` from `int_nvvm_tcgen05_ld_16x128b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x32.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x4_pack16` | `nvvm.tcgen05_ld_16x128b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x4` from `int_nvvm_tcgen05_ld_16x128b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x4.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x4_raw` | `nvvm.tcgen05_ld_16x128b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x4` from `int_nvvm_tcgen05_ld_16x128b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x4.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x64_pack16` | `nvvm.tcgen05_ld_16x128b_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x64` from `int_nvvm_tcgen05_ld_16x128b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x64.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x64_raw` | `nvvm.tcgen05_ld_16x128b_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x64` from `int_nvvm_tcgen05_ld_16x128b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x64.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x8_pack16` | `nvvm.tcgen05_ld_16x128b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x8` from `int_nvvm_tcgen05_ld_16x128b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x8.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x8_raw` | `nvvm.tcgen05_ld_16x128b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x8` from `int_nvvm_tcgen05_ld_16x128b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x128b.x8.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_pure` | `nvvm.tcgen05_ld_16x256b_pure` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x1` from `int_nvvm_tcgen05_ld_16x256b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x1.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x16_pack16` | `nvvm.tcgen05_ld_16x256b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x16` from `int_nvvm_tcgen05_ld_16x256b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x16.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x16_raw` | `nvvm.tcgen05_ld_16x256b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x16` from `int_nvvm_tcgen05_ld_16x256b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x16.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x1_pack16` | `nvvm.tcgen05_ld_16x256b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x1` from `int_nvvm_tcgen05_ld_16x256b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x1.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x1_raw` | `nvvm.tcgen05_ld_16x256b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x1` from `int_nvvm_tcgen05_ld_16x256b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x1.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x2_pack16` | `nvvm.tcgen05_ld_16x256b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x2` from `int_nvvm_tcgen05_ld_16x256b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x2.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x2_raw` | `nvvm.tcgen05_ld_16x256b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x2` from `int_nvvm_tcgen05_ld_16x256b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x2.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x32_pack16` | `nvvm.tcgen05_ld_16x256b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x32` from `int_nvvm_tcgen05_ld_16x256b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x32.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x32_raw` | `nvvm.tcgen05_ld_16x256b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x32` from `int_nvvm_tcgen05_ld_16x256b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x32.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x4_pack16` | `nvvm.tcgen05_ld_16x256b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x4` from `int_nvvm_tcgen05_ld_16x256b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x4.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x4_raw` | `nvvm.tcgen05_ld_16x256b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x4` from `int_nvvm_tcgen05_ld_16x256b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x4.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x8_pack16` | `nvvm.tcgen05_ld_16x256b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x8` from `int_nvvm_tcgen05_ld_16x256b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x8.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x8_pure` | `nvvm.tcgen05_ld_16x256b_x8_pure` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x8` from `int_nvvm_tcgen05_ld_16x256b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x8.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x8_raw` | `nvvm.tcgen05_ld_16x256b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x8` from `int_nvvm_tcgen05_ld_16x256b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x256b.x8.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x128_pack16` | `nvvm.tcgen05_ld_16x32bx2_x128_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x128` from `int_nvvm_tcgen05_ld_16x32bx2_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x128.pack::16b.b32 <register-list:128>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x128_raw` | `nvvm.tcgen05_ld_16x32bx2_x128_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x128` from `int_nvvm_tcgen05_ld_16x32bx2_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x128.b32 <register-list:128>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x16_pack16` | `nvvm.tcgen05_ld_16x32bx2_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x16` from `int_nvvm_tcgen05_ld_16x32bx2_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x16.pack::16b.b32 <register-list:16>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x16_raw` | `nvvm.tcgen05_ld_16x32bx2_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x16` from `int_nvvm_tcgen05_ld_16x32bx2_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x16.b32 <register-list:16>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x1_pack16` | `nvvm.tcgen05_ld_16x32bx2_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x1` from `int_nvvm_tcgen05_ld_16x32bx2_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x1.pack::16b.b32 <register-list:1>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x1_raw` | `nvvm.tcgen05_ld_16x32bx2_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x1` from `int_nvvm_tcgen05_ld_16x32bx2_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x1.b32 <register-list:1>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x2_pack16` | `nvvm.tcgen05_ld_16x32bx2_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x2` from `int_nvvm_tcgen05_ld_16x32bx2_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x2.pack::16b.b32 <register-list:2>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x2_raw` | `nvvm.tcgen05_ld_16x32bx2_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x2` from `int_nvvm_tcgen05_ld_16x32bx2_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x2.b32 <register-list:2>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x32_pack16` | `nvvm.tcgen05_ld_16x32bx2_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x32` from `int_nvvm_tcgen05_ld_16x32bx2_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x32.pack::16b.b32 <register-list:32>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x32_raw` | `nvvm.tcgen05_ld_16x32bx2_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x32` from `int_nvvm_tcgen05_ld_16x32bx2_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x32.b32 <register-list:32>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x4_pack16` | `nvvm.tcgen05_ld_16x32bx2_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x4` from `int_nvvm_tcgen05_ld_16x32bx2_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x4.pack::16b.b32 <register-list:4>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x4_raw` | `nvvm.tcgen05_ld_16x32bx2_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x4` from `int_nvvm_tcgen05_ld_16x32bx2_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x4.b32 <register-list:4>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x64_pack16` | `nvvm.tcgen05_ld_16x32bx2_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x64` from `int_nvvm_tcgen05_ld_16x32bx2_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x64.pack::16b.b32 <register-list:64>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x64_raw` | `nvvm.tcgen05_ld_16x32bx2_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x64` from `int_nvvm_tcgen05_ld_16x32bx2_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x64.b32 <register-list:64>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x8_pack16` | `nvvm.tcgen05_ld_16x32bx2_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x8` from `int_nvvm_tcgen05_ld_16x32bx2_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x8.pack::16b.b32 <register-list:8>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x8_raw` | `nvvm.tcgen05_ld_16x32bx2_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x8` from `int_nvvm_tcgen05_ld_16x32bx2_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x32bx2.x8.b32 <register-list:8>, [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x128_pack16` | `nvvm.tcgen05_ld_16x64b_x128_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x128` from `int_nvvm_tcgen05_ld_16x64b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x128.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x128_raw` | `nvvm.tcgen05_ld_16x64b_x128_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x128` from `int_nvvm_tcgen05_ld_16x64b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x128.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x16_pack16` | `nvvm.tcgen05_ld_16x64b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x16` from `int_nvvm_tcgen05_ld_16x64b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x16.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x16_raw` | `nvvm.tcgen05_ld_16x64b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x16` from `int_nvvm_tcgen05_ld_16x64b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x16.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x1_pack16` | `nvvm.tcgen05_ld_16x64b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x1` from `int_nvvm_tcgen05_ld_16x64b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x1.pack::16b.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x1_raw` | `nvvm.tcgen05_ld_16x64b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x1` from `int_nvvm_tcgen05_ld_16x64b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x1.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x2_pack16` | `nvvm.tcgen05_ld_16x64b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x2` from `int_nvvm_tcgen05_ld_16x64b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x2.pack::16b.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x2_raw` | `nvvm.tcgen05_ld_16x64b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x2` from `int_nvvm_tcgen05_ld_16x64b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x2.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x32_pack16` | `nvvm.tcgen05_ld_16x64b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x32` from `int_nvvm_tcgen05_ld_16x64b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x32.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x32_raw` | `nvvm.tcgen05_ld_16x64b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x32` from `int_nvvm_tcgen05_ld_16x64b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x32.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x4_pack16` | `nvvm.tcgen05_ld_16x64b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x4` from `int_nvvm_tcgen05_ld_16x64b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x4.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x4_raw` | `nvvm.tcgen05_ld_16x64b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x4` from `int_nvvm_tcgen05_ld_16x64b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x4.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x64_pack16` | `nvvm.tcgen05_ld_16x64b_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x64` from `int_nvvm_tcgen05_ld_16x64b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x64.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x64_raw` | `nvvm.tcgen05_ld_16x64b_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x64` from `int_nvvm_tcgen05_ld_16x64b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x64.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x8_pack16` | `nvvm.tcgen05_ld_16x64b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x8` from `int_nvvm_tcgen05_ld_16x64b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x8.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x8_raw` | `nvvm.tcgen05_ld_16x64b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x8` from `int_nvvm_tcgen05_ld_16x64b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.16x64b.x8.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x128_pack16` | `nvvm.tcgen05_ld_32x32b_x128_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x128` from `int_nvvm_tcgen05_ld_32x32b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x128.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x128_raw` | `nvvm.tcgen05_ld_32x32b_x128_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x128` from `int_nvvm_tcgen05_ld_32x32b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x128.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x16_pack16` | `nvvm.tcgen05_ld_32x32b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x16` from `int_nvvm_tcgen05_ld_32x32b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x16.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x16_raw` | `nvvm.tcgen05_ld_32x32b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x16` from `int_nvvm_tcgen05_ld_32x32b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x16.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x1_pack16` | `nvvm.tcgen05_ld_32x32b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x1` from `int_nvvm_tcgen05_ld_32x32b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x1.pack::16b.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x1_raw` | `nvvm.tcgen05_ld_32x32b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x1` from `int_nvvm_tcgen05_ld_32x32b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x1.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x2_pack16` | `nvvm.tcgen05_ld_32x32b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x2` from `int_nvvm_tcgen05_ld_32x32b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x2.pack::16b.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x2_raw` | `nvvm.tcgen05_ld_32x32b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x2` from `int_nvvm_tcgen05_ld_32x32b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x2.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x32_pack16` | `nvvm.tcgen05_ld_32x32b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x32` from `int_nvvm_tcgen05_ld_32x32b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x32.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x32_raw` | `nvvm.tcgen05_ld_32x32b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x32` from `int_nvvm_tcgen05_ld_32x32b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x32.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x4_pack16` | `nvvm.tcgen05_ld_32x32b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x4` from `int_nvvm_tcgen05_ld_32x32b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x4.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x4_raw` | `nvvm.tcgen05_ld_32x32b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x4` from `int_nvvm_tcgen05_ld_32x32b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x4.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x64_pack16` | `nvvm.tcgen05_ld_32x32b_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x64` from `int_nvvm_tcgen05_ld_32x32b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x64.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x64_raw` | `nvvm.tcgen05_ld_32x32b_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x64` from `int_nvvm_tcgen05_ld_32x32b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x64.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x8_pack16` | `nvvm.tcgen05_ld_32x32b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x8` from `int_nvvm_tcgen05_ld_32x32b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x8.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x8_raw` | `nvvm.tcgen05_ld_32x32b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x8` from `int_nvvm_tcgen05_ld_32x32b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.ld.sync.aligned.32x32b.x8.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_load_wait` | `nvvm.tcgen05_load_wait` | LLVM `llvm.nvvm.tcgen05.wait.ld` from `int_nvvm_tcgen05_wait_ld` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.wait::ld.sync.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_e2m1` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_e2m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_e3m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_e4m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_e5m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_f16` | `nvvm.tcgen05_mma_f16` | LLVM `llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1` from `int_nvvm_tcgen05_mma_shared_disable_output_lane_cg1` | unsafe; scope thread; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.mma.cta_group::1.kind::f16 [<address>], <register>, <register>, <register>, {%z, %z, %z, %z}, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_f16_cg2` | `nvvm.tcgen05_mma_f16_cg2` | LLVM `llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2` from `int_nvvm_tcgen05_mma_shared_disable_output_lane_cg2` | unsafe; scope thread; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.mma.cta_group::2.kind::f16 [<address>], <register>, <register>, <register>, {%z, %z, %z, %z, %z, %z, %z, %z}, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_sp_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.sp.shared` from `int_nvvm_tcgen05_mma_sp_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [<address>], <register>, <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_sp_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.sp.tensor` from `int_nvvm_tcgen05_mma_sp_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [<address>], [<address>], <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_sp_tensor_ashift` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.sp.tensor.ashift` from `int_nvvm_tcgen05_mma_sp_tensor_ashift` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.ashift [<address>], [<address>], <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.tensor` from `int_nvvm_tcgen05_mma_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_tensor_ashift` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.tensor.ashift` from `int_nvvm_tcgen05_mma_tensor_ashift` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.ashift [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_bf16` | `nvvm.tcgen05_mma_ws_bf16` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.mma.ws.cta_group::1.kind::f16 [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e2m1` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e2m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e3m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e4m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e5m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_f16` | `nvvm.tcgen05_mma_ws_f16` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.mma.ws.cta_group::1.kind::f16 [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.shared` from `int_nvvm_tcgen05_mma_ws_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_shared_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.shared.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_shared_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, <register>, %enable_pred, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.shared` from `int_nvvm_tcgen05_mma_ws_sp_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_shared_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.shared.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_sp_shared_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, [<address>], <register>, %enable_pred, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.tensor` from `int_nvvm_tcgen05_mma_ws_sp_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_tensor_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.tensor.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_sp_tensor_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, [<address>], <register>, %enable_pred, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_tensor_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_tensor_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.0-16696adc-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_tf32` | `nvvm.tcgen05_mma_ws_tf32` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.mma.ws.cta_group::1.kind::tf32 [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_relinquish_alloc_permit` | `nvvm.tcgen05_relinquish_alloc_permit` | LLVM `llvm.nvvm.tcgen05.relinq.alloc.permit.cg1` from `int_nvvm_tcgen05_relinq_alloc_permit_cg1` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.relinquish_alloc_permit.cta_group::1.sync.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_relinquish_alloc_permit_cg2` | `nvvm.tcgen05_relinquish_alloc_permit_cg2` | LLVM `llvm.nvvm.tcgen05.relinq.alloc.permit.cg2` from `int_nvvm_tcgen05_relinq_alloc_permit_cg2` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.relinquish_alloc_permit.cta_group::2.sync.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_shift_down` | `nvvm.tcgen05_shift_down` | LLVM `llvm.nvvm.tcgen05.shift.down.cg1` from `int_nvvm_tcgen05_shift_down_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.shift.cta_group::1.down [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_shift_down_cg2` | `nvvm.tcgen05_shift_down_cg2` | LLVM `llvm.nvvm.tcgen05.shift.down.cg2` from `int_nvvm_tcgen05_shift_down_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.shift.cta_group::2.down [<address>];` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x16_raw` | `nvvm.tcgen05_st_16x128b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x16` from `int_nvvm_tcgen05_st_16x128b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x16.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x16_unpack16` | `nvvm.tcgen05_st_16x128b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x16` from `int_nvvm_tcgen05_st_16x128b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x16.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x1_raw` | `nvvm.tcgen05_st_16x128b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x1` from `int_nvvm_tcgen05_st_16x128b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x1.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x1_unpack16` | `nvvm.tcgen05_st_16x128b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x1` from `int_nvvm_tcgen05_st_16x128b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x1.unpack::16b.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x2_raw` | `nvvm.tcgen05_st_16x128b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x2` from `int_nvvm_tcgen05_st_16x128b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x2.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x2_unpack16` | `nvvm.tcgen05_st_16x128b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x2` from `int_nvvm_tcgen05_st_16x128b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x2.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x32_raw` | `nvvm.tcgen05_st_16x128b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x32` from `int_nvvm_tcgen05_st_16x128b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x32.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x32_unpack16` | `nvvm.tcgen05_st_16x128b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x32` from `int_nvvm_tcgen05_st_16x128b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x32.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x4_raw` | `nvvm.tcgen05_st_16x128b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x4` from `int_nvvm_tcgen05_st_16x128b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x4.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x4_unpack16` | `nvvm.tcgen05_st_16x128b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x4` from `int_nvvm_tcgen05_st_16x128b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x4.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x64_raw` | `nvvm.tcgen05_st_16x128b_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x64` from `int_nvvm_tcgen05_st_16x128b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x64.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x64_unpack16` | `nvvm.tcgen05_st_16x128b_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x64` from `int_nvvm_tcgen05_st_16x128b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x64.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x8_raw` | `nvvm.tcgen05_st_16x128b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x8` from `int_nvvm_tcgen05_st_16x128b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x8.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x8_unpack16` | `nvvm.tcgen05_st_16x128b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x8` from `int_nvvm_tcgen05_st_16x128b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x128b.x8.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x16_raw` | `nvvm.tcgen05_st_16x256b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x16` from `int_nvvm_tcgen05_st_16x256b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x16.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x16_unpack16` | `nvvm.tcgen05_st_16x256b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x16` from `int_nvvm_tcgen05_st_16x256b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x16.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x1_raw` | `nvvm.tcgen05_st_16x256b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x1` from `int_nvvm_tcgen05_st_16x256b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x1.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x1_unpack16` | `nvvm.tcgen05_st_16x256b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x1` from `int_nvvm_tcgen05_st_16x256b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x1.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x2_raw` | `nvvm.tcgen05_st_16x256b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x2` from `int_nvvm_tcgen05_st_16x256b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x2.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x2_unpack16` | `nvvm.tcgen05_st_16x256b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x2` from `int_nvvm_tcgen05_st_16x256b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x2.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x32_raw` | `nvvm.tcgen05_st_16x256b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x32` from `int_nvvm_tcgen05_st_16x256b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x32.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x32_unpack16` | `nvvm.tcgen05_st_16x256b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x32` from `int_nvvm_tcgen05_st_16x256b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x32.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x4_raw` | `nvvm.tcgen05_st_16x256b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x4` from `int_nvvm_tcgen05_st_16x256b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x4.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x4_unpack16` | `nvvm.tcgen05_st_16x256b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x4` from `int_nvvm_tcgen05_st_16x256b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x4.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x8_raw` | `nvvm.tcgen05_st_16x256b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x8` from `int_nvvm_tcgen05_st_16x256b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x8.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x8_unpack16` | `nvvm.tcgen05_st_16x256b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x8` from `int_nvvm_tcgen05_st_16x256b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x256b.x8.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x128_raw` | `nvvm.tcgen05_st_16x32bx2_x128_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x128` from `int_nvvm_tcgen05_st_16x32bx2_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x128.b32 [<address>], <immediate>, <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x128_unpack16` | `nvvm.tcgen05_st_16x32bx2_x128_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x128` from `int_nvvm_tcgen05_st_16x32bx2_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x128.unpack::16b.b32 [<address>], <immediate>, <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x16_raw` | `nvvm.tcgen05_st_16x32bx2_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x16` from `int_nvvm_tcgen05_st_16x32bx2_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x16.b32 [<address>], <immediate>, <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x16_unpack16` | `nvvm.tcgen05_st_16x32bx2_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x16` from `int_nvvm_tcgen05_st_16x32bx2_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x16.unpack::16b.b32 [<address>], <immediate>, <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x1_raw` | `nvvm.tcgen05_st_16x32bx2_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x1` from `int_nvvm_tcgen05_st_16x32bx2_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x1.b32 [<address>], <immediate>, <register-list:1>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x1_unpack16` | `nvvm.tcgen05_st_16x32bx2_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x1` from `int_nvvm_tcgen05_st_16x32bx2_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x1.unpack::16b.b32 [<address>], <immediate>, <register-list:1>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x2_raw` | `nvvm.tcgen05_st_16x32bx2_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x2` from `int_nvvm_tcgen05_st_16x32bx2_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x2.b32 [<address>], <immediate>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x2_unpack16` | `nvvm.tcgen05_st_16x32bx2_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x2` from `int_nvvm_tcgen05_st_16x32bx2_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x2.unpack::16b.b32 [<address>], <immediate>, <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x32_raw` | `nvvm.tcgen05_st_16x32bx2_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x32` from `int_nvvm_tcgen05_st_16x32bx2_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x32.b32 [<address>], <immediate>, <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x32_unpack16` | `nvvm.tcgen05_st_16x32bx2_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x32` from `int_nvvm_tcgen05_st_16x32bx2_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x32.unpack::16b.b32 [<address>], <immediate>, <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x4_raw` | `nvvm.tcgen05_st_16x32bx2_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x4` from `int_nvvm_tcgen05_st_16x32bx2_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x4.b32 [<address>], <immediate>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x4_unpack16` | `nvvm.tcgen05_st_16x32bx2_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x4` from `int_nvvm_tcgen05_st_16x32bx2_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x4.unpack::16b.b32 [<address>], <immediate>, <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x64_raw` | `nvvm.tcgen05_st_16x32bx2_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x64` from `int_nvvm_tcgen05_st_16x32bx2_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x64.b32 [<address>], <immediate>, <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x64_unpack16` | `nvvm.tcgen05_st_16x32bx2_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x64` from `int_nvvm_tcgen05_st_16x32bx2_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x64.unpack::16b.b32 [<address>], <immediate>, <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x8_raw` | `nvvm.tcgen05_st_16x32bx2_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x8` from `int_nvvm_tcgen05_st_16x32bx2_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x8.b32 [<address>], <immediate>, <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x8_unpack16` | `nvvm.tcgen05_st_16x32bx2_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x8` from `int_nvvm_tcgen05_st_16x32bx2_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x32bx2.x8.unpack::16b.b32 [<address>], <immediate>, <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x128_raw` | `nvvm.tcgen05_st_16x64b_x128_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x128` from `int_nvvm_tcgen05_st_16x64b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x128.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x128_unpack16` | `nvvm.tcgen05_st_16x64b_x128_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x128` from `int_nvvm_tcgen05_st_16x64b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x128.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x16_raw` | `nvvm.tcgen05_st_16x64b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x16` from `int_nvvm_tcgen05_st_16x64b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x16.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x16_unpack16` | `nvvm.tcgen05_st_16x64b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x16` from `int_nvvm_tcgen05_st_16x64b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x16.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x1_raw` | `nvvm.tcgen05_st_16x64b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x1` from `int_nvvm_tcgen05_st_16x64b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x1.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x1_unpack16` | `nvvm.tcgen05_st_16x64b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x1` from `int_nvvm_tcgen05_st_16x64b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x1.unpack::16b.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x2_raw` | `nvvm.tcgen05_st_16x64b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x2` from `int_nvvm_tcgen05_st_16x64b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x2.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x2_unpack16` | `nvvm.tcgen05_st_16x64b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x2` from `int_nvvm_tcgen05_st_16x64b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x2.unpack::16b.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x32_raw` | `nvvm.tcgen05_st_16x64b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x32` from `int_nvvm_tcgen05_st_16x64b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x32.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x32_unpack16` | `nvvm.tcgen05_st_16x64b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x32` from `int_nvvm_tcgen05_st_16x64b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x32.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x4_raw` | `nvvm.tcgen05_st_16x64b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x4` from `int_nvvm_tcgen05_st_16x64b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x4.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x4_unpack16` | `nvvm.tcgen05_st_16x64b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x4` from `int_nvvm_tcgen05_st_16x64b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x4.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x64_raw` | `nvvm.tcgen05_st_16x64b_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x64` from `int_nvvm_tcgen05_st_16x64b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x64.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x64_unpack16` | `nvvm.tcgen05_st_16x64b_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x64` from `int_nvvm_tcgen05_st_16x64b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x64.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x8_raw` | `nvvm.tcgen05_st_16x64b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x8` from `int_nvvm_tcgen05_st_16x64b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x8.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x8_unpack16` | `nvvm.tcgen05_st_16x64b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x8` from `int_nvvm_tcgen05_st_16x64b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.16x64b.x8.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x128_raw` | `nvvm.tcgen05_st_32x32b_x128_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x128` from `int_nvvm_tcgen05_st_32x32b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x128.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x128_unpack16` | `nvvm.tcgen05_st_32x32b_x128_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x128` from `int_nvvm_tcgen05_st_32x32b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x128.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x16_raw` | `nvvm.tcgen05_st_32x32b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x16` from `int_nvvm_tcgen05_st_32x32b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x16.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x16_unpack16` | `nvvm.tcgen05_st_32x32b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x16` from `int_nvvm_tcgen05_st_32x32b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x16.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x1_raw` | `nvvm.tcgen05_st_32x32b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x1` from `int_nvvm_tcgen05_st_32x32b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x1.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x1_unpack16` | `nvvm.tcgen05_st_32x32b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x1` from `int_nvvm_tcgen05_st_32x32b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x1.unpack::16b.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x2_raw` | `nvvm.tcgen05_st_32x32b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x2` from `int_nvvm_tcgen05_st_32x32b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x2.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x2_unpack16` | `nvvm.tcgen05_st_32x32b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x2` from `int_nvvm_tcgen05_st_32x32b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x2.unpack::16b.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x32_raw` | `nvvm.tcgen05_st_32x32b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x32` from `int_nvvm_tcgen05_st_32x32b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x32.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x32_unpack16` | `nvvm.tcgen05_st_32x32b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x32` from `int_nvvm_tcgen05_st_32x32b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x32.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x4_raw` | `nvvm.tcgen05_st_32x32b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x4` from `int_nvvm_tcgen05_st_32x32b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x4.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x4_unpack16` | `nvvm.tcgen05_st_32x32b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x4` from `int_nvvm_tcgen05_st_32x32b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x4.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x64_raw` | `nvvm.tcgen05_st_32x32b_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x64` from `int_nvvm_tcgen05_st_32x32b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x64.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x64_unpack16` | `nvvm.tcgen05_st_32x32b_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x64` from `int_nvvm_tcgen05_st_32x32b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x64.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x8_raw` | `nvvm.tcgen05_st_32x32b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x8` from `int_nvvm_tcgen05_st_32x32b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x8.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x8_unpack16` | `nvvm.tcgen05_st_32x32b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x8` from `int_nvvm_tcgen05_st_32x32b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.st.sync.aligned.32x32b.x8.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tcgen05::tcgen05_store_wait` | `nvvm.tcgen05_store_wait` | LLVM `llvm.nvvm.tcgen05.wait.st` from `int_nvvm_tcgen05_wait_st` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.0-16696adc`; expects `tcgen05.wait::st.sync.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_box_dim` | `nvvm.tensormap_replace_box_dim` | LLVM `llvm.nvvm.tensormap.replace.box.dim` from `int_nvvm_tensormap_replace_box_dim` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.box_dim.global.b1024.b32 [<address>], <immediate>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_element_stride` | `nvvm.tensormap_replace_element_stride` | LLVM `llvm.nvvm.tensormap.replace.element.stride` from `int_nvvm_tensormap_replace_element_stride` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.element_stride.global.b1024.b32 [<address>], <immediate>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_element_type` | `nvvm.tensormap_replace_element_type` | LLVM `llvm.nvvm.tensormap.replace.elemtype` from `int_nvvm_tensormap_replace_elemtype` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.elemtype.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_fill_mode` | `nvvm.tensormap_replace_fill_mode` | LLVM `llvm.nvvm.tensormap.replace.fill.mode` from `int_nvvm_tensormap_replace_fill_mode` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.fill_mode.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_global_address` | `nvvm.tensormap_replace_global_address` | LLVM `llvm.nvvm.tensormap.replace.global.address` from `int_nvvm_tensormap_replace_global_address` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.global_address.global.b1024.b64 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_global_dim` | `nvvm.tensormap_replace_global_dim` | LLVM `llvm.nvvm.tensormap.replace.global.dim` from `int_nvvm_tensormap_replace_global_dim` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.global_dim.global.b1024.b32 [<address>], <immediate>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_global_stride` | `nvvm.tensormap_replace_global_stride` | LLVM `llvm.nvvm.tensormap.replace.global.stride` from `int_nvvm_tensormap_replace_global_stride` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.global_stride.global.b1024.b64 [<address>], <immediate>, <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_interleave_layout` | `nvvm.tensormap_replace_interleave_layout` | LLVM `llvm.nvvm.tensormap.replace.interleave.layout` from `int_nvvm_tensormap_replace_interleave_layout` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.interleave_layout.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_rank` | `nvvm.tensormap_replace_rank` | LLVM `llvm.nvvm.tensormap.replace.rank` from `int_nvvm_tensormap_replace_rank` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.rank.global.b1024.b32 [<address>], <register>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_swizzle_atomicity` | `nvvm.tensormap_replace_swizzle_atomicity` | LLVM `llvm.nvvm.tensormap.replace.swizzle.atomicity` from `int_nvvm_tensormap_replace_swizzle_atomicity` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.swizzle_atomicity.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::tma::tensormap_replace_swizzle_mode` | `nvvm.tensormap_replace_swizzle_mode` | LLVM `llvm.nvvm.tensormap.replace.swizzle.mode` from `int_nvvm_tensormap_replace_swizzle_mode` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.0-16696adc-tma`; expects `tensormap.replace.tile.swizzle_mode.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::thread_idx_x` | `nvvm.read_ptx_sreg_tid_x` | LLVM `llvm.nvvm.read.ptx.sreg.tid.x` from `int_nvvm_read_ptx_sreg_tid_x` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.1 Special Registers: %tid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %tid.x;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::thread_idx_y` | `nvvm.read_ptx_sreg_tid_y` | LLVM `llvm.nvvm.read.ptx.sreg.tid.y` from `int_nvvm_read_ptx_sreg_tid_y` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.1 Special Registers: %tid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %tid.y;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::sreg::thread_idx_z` | `nvvm.read_ptx_sreg_tid_z` | LLVM `llvm.nvvm.read.ptx.sreg.tid.z` from `int_nvvm_read_ptx_sreg_tid_z` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.1 Special Registers: %tid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %tid.z;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::fence::threadfence` | `nvvm.threadfence` | LLVM `llvm.nvvm.membar.gl` from `int_nvvm_membar_gl` | safe; scope device; impure; memory read_write; convergent false | PTX 1.4 on all ([PTX ISA 9.3, 9.7.14.4 Parallel Synchronization and Communication Instructions: membar / fence](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence)) | validated on `rust-llvm-23.1.0-16696adc`; expects `membar.gl;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::fence::threadfence_block` | `nvvm.threadfence_block` | LLVM `llvm.nvvm.membar.cta` from `int_nvvm_membar_cta` | safe; scope cta; impure; memory read_write; convergent false | PTX 1.4 on all ([PTX ISA 9.3, 9.7.14.4 Parallel Synchronization and Communication Instructions: membar / fence](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence)) | validated on `rust-llvm-23.1.0-16696adc`; expects `membar.cta;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::fence::threadfence_system` | `nvvm.threadfence_system` | LLVM `llvm.nvvm.membar.sys` from `int_nvvm_membar_sys` | safe; scope system; impure; memory read_write; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.14.4 Parallel Synchronization and Communication Instructions: membar / fence](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence)) | validated on `rust-llvm-23.1.0-16696adc`; expects `membar.sys;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::shared::total_smem_size` | `nvvm.read_ptx_sreg_total_smem_size` | LLVM `llvm.nvvm.read.ptx.sreg.total_smem_size` from `int_nvvm_read_ptx_sreg_total_smem_size` | safe; scope cta; pure; memory none; convergent false | PTX 4.1 on sm_20+ ([PTX ISA 9.3, 10.30 Special Registers: %total_smem_size](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-total-smem-size)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %total_smem_size;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::debug::trap` | `nvvm.trap` | PTX-native `trap` | safe; scope thread; impure; memory none; convergent false | PTX 1.0 on all ([PTX ISA 9.3, 9.7.20.4 Miscellaneous Instructions: trap](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-trap)) | validated on `rust-llvm-23.1.0-16696adc`; expects `trap;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::uni_sync` | `nvvm.vote_sync_uni` | LLVM `llvm.nvvm.vote.uni.sync` from `int_nvvm_vote_uni_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.0-16696adc`; expects `vote.sync.uni.pred <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::warp::warpid` | `nvvm.read_ptx_sreg_warpid` | LLVM `llvm.nvvm.read.ptx.sreg.warpid` from `int_nvvm_read_ptx_sreg_warpid` | safe; scope cta; impure; memory none; convergent false | PTX 1.3 on all ([PTX ISA 9.3, 10.4 Special Registers: %warpid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-warpid)) | validated on `rust-llvm-23.1.0-16696adc`; expects `mov.u32 <register>, %warpid;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::wgmma::wgmma_commit_group` | `nvvm.wgmma_commit_group_sync_aligned` | LLVM `llvm.nvvm.wgmma.commit_group.sync.aligned` from `int_nvvm_wgmma_commit_group_sync_aligned` | unsafe; scope warpgroup; impure; memory read_write; convergent true | PTX 8.0 on sm_90a ([PTX ISA 9.3, Asynchronous Warpgroup Level Matrix Instructions: WGMMA control](https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions)) | validated on `rust-llvm-23.1.0-16696adc`; expects `wgmma.commit_group.sync.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::wgmma::wgmma_fence` | `nvvm.wgmma_fence_sync_aligned` | LLVM `llvm.nvvm.wgmma.fence.sync.aligned` from `int_nvvm_wgmma_fence_sync_aligned` | unsafe; scope warpgroup; impure; memory read_write; convergent true | PTX 8.0 on sm_90a ([PTX ISA 9.3, Asynchronous Warpgroup Level Matrix Instructions: WGMMA control](https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions)) | validated on `rust-llvm-23.1.0-16696adc`; expects `wgmma.fence.sync.aligned;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
-| `cuda_intrinsics::wgmma::wgmma_wait_group` | `nvvm.wgmma_wait_group_sync_aligned` | LLVM `llvm.nvvm.wgmma.wait_group.sync.aligned` from `int_nvvm_wgmma_wait_group_sync_aligned` | unsafe; scope warpgroup; impure; memory read_write; convergent true | PTX 8.0 on sm_90a ([PTX ISA 9.3, Asynchronous Warpgroup Level Matrix Instructions: WGMMA control](https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions)) | validated on `rust-llvm-23.1.0-16696adc`; expects `wgmma.wait_group.sync.aligned <immediate>;` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`) |
+| `cuda_intrinsics::bf16x2::abs_bf16x2` | `nvvm.abs_bf16x2` | LLVM `llvm.nvvm.fabs` from `int_nvvm_fabs` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.6 Half Precision Floating Point Instructions: abs](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `abs.bf16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::abs_f16x2` | `nvvm.abs_f16x2` | LLVM `llvm.nvvm.fabs` from `int_nvvm_fabs` | safe; scope thread; pure; memory none; convergent false | PTX 6.5 on sm_70+ ([PTX ISA 9.3, 9.7.4.6 Half Precision Floating Point Instructions: abs](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `abs.f16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::active_mask` | `nvvm.activemask` | LLVM `llvm.nvvm.activemask` from `int_nvvm_activemask` | safe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.2 on sm_30+ ([PTX ISA 9.3, 9.7.14.12 Parallel Synchronization and Communication Instructions: activemask](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-activemask)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `activemask.b32 <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::add_bf16x2` | `nvvm.add_bf16x2` | PTX-native `add.rn.bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.4.1 Half Precision Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::add_f16x2` | `nvvm.add_f16x2` | PTX-native `add.rn.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.1 Half Precision Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::add_f32x2` | `nvvm.add_f32x2` | PTX-native `add.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.1 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::add_ftz_f32x2` | `nvvm.add_ftz_f32x2` | PTX-native `add.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.1 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.ftz.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.f` from `int_nvvm_add_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rm.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.d` from `int_nvvm_add_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rm.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.ftz.f` from `int_nvvm_add_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rm.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rm_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.ftz.sat.f` from `int_nvvm_add_rm_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rm.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rm_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rm.sat.f` from `int_nvvm_add_rm_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rm.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.f` from `int_nvvm_add_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.d` from `int_nvvm_add_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.ftz.f` from `int_nvvm_add_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rn_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.ftz.sat.f` from `int_nvvm_add_rn_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rn_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rn.sat.f` from `int_nvvm_add_rn_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rn.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.f` from `int_nvvm_add_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rp.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.d` from `int_nvvm_add_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rp.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.ftz.f` from `int_nvvm_add_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rp.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rp_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.ftz.sat.f` from `int_nvvm_add_rp_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rp.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rp_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rp.sat.f` from `int_nvvm_add_rp_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rp.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.f` from `int_nvvm_add_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.d` from `int_nvvm_add_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rz.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.ftz.f` from `int_nvvm_add_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rz.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rz_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.ftz.sat.f` from `int_nvvm_add_rz_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rz.sat.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::add_rz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.add.rz.sat.f` from `int_nvvm_add_rz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: add](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `add.rz.sat.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::all_sync` | `nvvm.vote_sync_all` | LLVM `llvm.nvvm.vote.all.sync` from `int_nvvm_vote_all_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `vote.sync.all.pred <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::any_sync` | `nvvm.vote_sync_any` | LLVM `llvm.nvvm.vote.any.sync` from `int_nvvm_vote_any_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `vote.sync.any.pred <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::ballot_sync` | `nvvm.vote_sync_ballot` | LLVM `llvm.nvvm.vote.ballot.sync` from `int_nvvm_vote_ballot_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `vote.sync.ballot.b32 <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::barrier_cluster_arrive` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive` from `int_nvvm_barrier_cluster_arrive` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `barrier.cluster.arrive;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::barrier_cluster_arrive_aligned` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive.aligned` from `int_nvvm_barrier_cluster_arrive_aligned` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `barrier.cluster.arrive.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::barrier_cluster_arrive_relaxed` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive.relaxed` from `int_nvvm_barrier_cluster_arrive_relaxed` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `barrier.cluster.arrive.relaxed;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::barrier_cluster_arrive_relaxed_aligned` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.arrive.relaxed.aligned` from `int_nvvm_barrier_cluster_arrive_relaxed_aligned` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `barrier.cluster.arrive.relaxed.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::barrier_cluster_wait` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.wait` from `int_nvvm_barrier_cluster_wait` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `barrier.cluster.wait;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::barrier_cluster_wait_aligned` | `nvvm.cluster_barrier` | LLVM `llvm.nvvm.barrier.cluster.wait.aligned` from `int_nvvm_barrier_cluster_wait_aligned` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: barrier.cluster](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-barrier-cluster)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `barrier.cluster.wait.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::barrier_cta_arrive` | `nvvm.barrier_cta_arrive_count` | LLVM `llvm.nvvm.barrier.cta.arrive.count` from `int_nvvm_barrier_cta_arrive_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `barrier.arrive <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::barrier_cta_arrive_aligned` | `nvvm.barrier_cta_arrive_aligned_count` | LLVM `llvm.nvvm.barrier.cta.arrive.aligned.count` from `int_nvvm_barrier_cta_arrive_aligned_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `bar.arrive <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::barrier_cta_sync` | `nvvm.barrier_cta_sync_count` | LLVM `llvm.nvvm.barrier.cta.sync.count` from `int_nvvm_barrier_cta_sync_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `barrier.sync <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::barrier_cta_sync_aligned` | `nvvm.barrier_cta_sync_aligned_count` | LLVM `llvm.nvvm.barrier.cta.sync.aligned.count` from `int_nvvm_barrier_cta_sync_aligned_count` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `bar.sync <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::block_dim_x` | `nvvm.read_ptx_sreg_ntid_x` | LLVM `llvm.nvvm.read.ptx.sreg.ntid.x` from `int_nvvm_read_ptx_sreg_ntid_x` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.2 Special Registers: %ntid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %ntid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::block_dim_y` | `nvvm.read_ptx_sreg_ntid_y` | LLVM `llvm.nvvm.read.ptx.sreg.ntid.y` from `int_nvvm_read_ptx_sreg_ntid_y` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.2 Special Registers: %ntid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %ntid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::block_dim_z` | `nvvm.read_ptx_sreg_ntid_z` | LLVM `llvm.nvvm.read.ptx.sreg.ntid.z` from `int_nvvm_read_ptx_sreg_ntid_z` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.2 Special Registers: %ntid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ntid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %ntid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::block_idx_x` | `nvvm.read_ptx_sreg_ctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.ctaid.x` from `int_nvvm_read_ptx_sreg_ctaid_x` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.6 Special Registers: %ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %ctaid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::block_idx_y` | `nvvm.read_ptx_sreg_ctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.ctaid.y` from `int_nvvm_read_ptx_sreg_ctaid_y` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.6 Special Registers: %ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %ctaid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::block_idx_z` | `nvvm.read_ptx_sreg_ctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.ctaid.z` from `int_nvvm_read_ptx_sreg_ctaid_z` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.6 Special Registers: %ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-ctaid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %ctaid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::debug::breakpoint` | `nvvm.brkpt` | PTX-native `brkpt` | safe; scope thread; impure; memory none; convergent false | PTX 1.0 on sm_11+ ([PTX ISA 9.3, 9.7.20.1 Miscellaneous Instructions: brkpt](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-brkpt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `brkpt;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::clc::clc_query_get_first_ctaid_x` | `nvvm.clc_query_get_first_ctaid_x` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.get_first_ctaid.x` from `int_nvvm_clusterlaunchcontrol_query_cancel_get_first_ctaid_x` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `clusterlaunchcontrol.query_cancel.get_first_ctaid::x.b32.b128 <register>, %clc_handle;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::clc::clc_query_get_first_ctaid_y` | `nvvm.clc_query_get_first_ctaid_y` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.get_first_ctaid.y` from `int_nvvm_clusterlaunchcontrol_query_cancel_get_first_ctaid_y` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `clusterlaunchcontrol.query_cancel.get_first_ctaid::y.b32.b128 <register>, %clc_handle;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::clc::clc_query_get_first_ctaid_z` | `nvvm.clc_query_get_first_ctaid_z` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.get_first_ctaid.z` from `int_nvvm_clusterlaunchcontrol_query_cancel_get_first_ctaid_z` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `clusterlaunchcontrol.query_cancel.get_first_ctaid::z.b32.b128 <register>, %clc_handle;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::clc::clc_query_is_canceled` | `nvvm.clc_query_is_canceled` | LLVM `llvm.nvvm.clusterlaunchcontrol.query_cancel.is_canceled` from `int_nvvm_clusterlaunchcontrol_query_cancel_is_canceled` | unsafe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `clusterlaunchcontrol.query_cancel.is_canceled.pred.b128 <register>, %clc_handle;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::clc::clc_try_cancel` | `nvvm.clc_try_cancel` | LLVM `llvm.nvvm.clusterlaunchcontrol.try_cancel.async.shared` from `int_nvvm_clusterlaunchcontrol_try_cancel_async_shared` | unsafe; scope cta; impure; memory read_write; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `clusterlaunchcontrol.try_cancel.async.shared::cta.mbarrier::complete_tx::bytes.b128 [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::clc::clc_try_cancel_multicast` | `nvvm.clc_try_cancel_multicast` | LLVM `llvm.nvvm.clusterlaunchcontrol.try_cancel.async.multicast.shared` from `int_nvvm_clusterlaunchcontrol_try_cancel_async_multicast_shared` | unsafe; scope cta; impure; memory read_write; convergent false | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a or sm_120a or sm_121a ([PTX ISA 9.3, 9.7.14.18-19 Cluster Launch Control](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-clusterlaunchcontrol-try-cancel)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `clusterlaunchcontrol.try_cancel.async.shared::cta.mbarrier::complete_tx::bytes.multicast::cluster::all.b128 [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::debug::clock` | `nvvm.read_ptx_sreg_clock` | LLVM `llvm.nvvm.read.ptx.sreg.clock` from `int_nvvm_read_ptx_sreg_clock` | safe; scope sm; impure; memory inaccessible_read_write; convergent false | PTX 1.0 on all ([PTX ISA 9.3, 10.23 Special Registers: %clock, %clock_hi](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clock-clock-hi)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %clock;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::debug::clock64` | `nvvm.read_ptx_sreg_clock64` | LLVM `llvm.nvvm.read.ptx.sreg.clock64` from `int_nvvm_read_ptx_sreg_clock64` | safe; scope sm; impure; memory inaccessible_read_write; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.24 Special Registers: %clock64](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clock64)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u64 <register>, %clock64;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_block_count` | `nvvm.read_ptx_sreg_cluster_nctarank` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctarank` from `int_nvvm_read_ptx_sreg_cluster_nctarank` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.17 Special Registers: %cluster_nctarank](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctarank)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_nctarank;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_block_idx_x` | `nvvm.read_ptx_sreg_cluster_ctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctaid.x` from `int_nvvm_read_ptx_sreg_cluster_ctaid_x` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.14 Special Registers: %cluster_ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_ctaid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_block_idx_y` | `nvvm.read_ptx_sreg_cluster_ctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctaid.y` from `int_nvvm_read_ptx_sreg_cluster_ctaid_y` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.14 Special Registers: %cluster_ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_ctaid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_block_idx_z` | `nvvm.read_ptx_sreg_cluster_ctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctaid.z` from `int_nvvm_read_ptx_sreg_cluster_ctaid_z` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.14 Special Registers: %cluster_ctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctaid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_ctaid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_block_rank` | `nvvm.read_ptx_sreg_cluster_ctarank` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.ctarank` from `int_nvvm_read_ptx_sreg_cluster_ctarank` | safe; scope cta; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.16 Special Registers: %cluster_ctarank](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-ctarank)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_ctarank;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_dim_x` | `nvvm.read_ptx_sreg_cluster_nctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctaid.x` from `int_nvvm_read_ptx_sreg_cluster_nctaid_x` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.15 Special Registers: %cluster_nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_nctaid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_dim_y` | `nvvm.read_ptx_sreg_cluster_nctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctaid.y` from `int_nvvm_read_ptx_sreg_cluster_nctaid_y` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.15 Special Registers: %cluster_nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_nctaid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_dim_z` | `nvvm.read_ptx_sreg_cluster_nctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.cluster.nctaid.z` from `int_nvvm_read_ptx_sreg_cluster_nctaid_z` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.15 Special Registers: %cluster_nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-cluster-nctaid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %cluster_nctaid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_grid_dim_x` | `nvvm.read_ptx_sreg_nclusterid_x` | LLVM `llvm.nvvm.read.ptx.sreg.nclusterid.x` from `int_nvvm_read_ptx_sreg_nclusterid_x` | safe; scope grid; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.13 Special Registers: %nclusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nclusterid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_grid_dim_y` | `nvvm.read_ptx_sreg_nclusterid_y` | LLVM `llvm.nvvm.read.ptx.sreg.nclusterid.y` from `int_nvvm_read_ptx_sreg_nclusterid_y` | safe; scope grid; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.13 Special Registers: %nclusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nclusterid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_grid_dim_z` | `nvvm.read_ptx_sreg_nclusterid_z` | LLVM `llvm.nvvm.read.ptx.sreg.nclusterid.z` from `int_nvvm_read_ptx_sreg_nclusterid_z` | safe; scope grid; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.13 Special Registers: %nclusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nclusterid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nclusterid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_idx_x` | `nvvm.read_ptx_sreg_clusterid_x` | LLVM `llvm.nvvm.read.ptx.sreg.clusterid.x` from `int_nvvm_read_ptx_sreg_clusterid_x` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.12 Special Registers: %clusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %clusterid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_idx_y` | `nvvm.read_ptx_sreg_clusterid_y` | LLVM `llvm.nvvm.read.ptx.sreg.clusterid.y` from `int_nvvm_read_ptx_sreg_clusterid_y` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.12 Special Registers: %clusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %clusterid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::cluster_idx_z` | `nvvm.read_ptx_sreg_clusterid_z` | LLVM `llvm.nvvm.read.ptx.sreg.clusterid.z` from `int_nvvm_read_ptx_sreg_clusterid_z` | safe; scope cluster; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 10.12 Special Registers: %clusterid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-clusterid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %clusterid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::cos_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.cos.approx.f` from `int_nvvm_cos_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.11 Floating Point Instructions: cos](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-cos)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cos.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::cos_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.cos.approx.ftz.f` from `int_nvvm_cos_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.11 Floating Point Instructions: cos](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-cos)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cos.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_commit_group` | `nvvm.cp_async_bulk_commit_group` | LLVM `llvm.nvvm.cp.async.bulk.commit.group` from `int_nvvm_cp_async_bulk_commit_group` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.commit_group;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_1d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_1d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_1d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.1d.L2.global.tile [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_1d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_1d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_1d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.1d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_2d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_2d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_2d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_2d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_3d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_3d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_3d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.3d.L2.global.tile [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_3d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_3d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_3d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.3d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_4d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_4d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_4d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.4d.L2.global.tile [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_4d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_4d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_4d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.4d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_5d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_5d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_5d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.5d.L2.global.tile [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_5d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_5d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_5d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.5d.L2.global.tile.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_gather4_2d_l2` | `nvvm.cp_async_bulk_prefetch_tensor_gather4_2d_l2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.gather4.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_gather4_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile::gather4 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint` | `nvvm.cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint` | LLVM `llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.gather4.2d` from `int_nvvm_cp_async_bulk_tensor_prefetch_tile_gather4_2d` | unsafe; scope thread; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.prefetch.tensor.2d.L2.global.tile::gather4.L2::cache_hint [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_1d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.1d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.1d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_1d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.1d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.1d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.2d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_g2s_multicast` | `nvvm.cp_async_bulk_tensor_g2s_tile_2d_multicast` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.2d.shared::cluster.global.tile.mbarrier::complete_tx::bytes.multicast::cluster [<address>], [<address>], [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_g2s_multicast_cg2` | `nvvm.cp_async_bulk_tensor_g2s_tile_2d_multicast_cg2` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.2d.shared::cluster.global.tile.mbarrier::complete_tx::bytes.multicast::cluster.cta_group::2 [<address>], [<address>], [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_2d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.2d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.2d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_3d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.3d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.3d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_3d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.3d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.3d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_4d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.4d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.4d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_4d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.4d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.4d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_5d_g2s` | `nvvm.cp_async_bulk_tensor_g2s_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.g2s.tile.5d` from `int_nvvm_cp_async_bulk_tensor_g2s_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.5d.shared::cluster.global.tile.mbarrier::complete_tx::bytes [<address>], [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_5d_s2g` | `nvvm.cp_async_bulk_tensor_s2g_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.s2g.tile.5d` from `int_nvvm_cp_async_bulk_tensor_s2g_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.tensor.5d.global.shared::cta.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_add_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.add.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_add_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.add.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_add_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.add.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_add_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_add_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.add.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_add_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.add.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_and_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.and.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_and_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.and.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_and_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.and.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_and_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_and_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.and.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_and_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.and.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_dec_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.dec.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_dec_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.dec.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_dec_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.dec.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_dec_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_dec_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.dec.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_dec_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.dec.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_inc_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.inc.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_inc_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.inc.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_inc_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.inc.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_inc_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_inc_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.inc.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_inc_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.inc.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_max_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.max.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_max_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.max.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_max_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.max.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_max_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_max_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.max.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_max_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.max.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_min_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.min.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_min_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.min.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_min_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.min.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_min_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_min_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.min.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_min_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.min.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_or_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.or.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_or_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.or.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_or_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.or.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_or_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_or_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.or.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_or_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.or.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_im2col_3d` | `nvvm.cp_async_bulk_tensor_reduce_xor_im2col_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.im2col.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_im2col_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.xor.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_im2col_4d` | `nvvm.cp_async_bulk_tensor_reduce_xor_im2col_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.im2col.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_im2col_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.xor.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_im2col_5d` | `nvvm.cp_async_bulk_tensor_reduce_xor_im2col_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.im2col.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_im2col_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.xor.im2col_no_offs.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_1d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_1d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.1d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_1d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.1d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_2d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_2d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.2d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_2d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.2d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_3d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_3d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.3d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_3d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.3d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_4d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_4d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.4d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_4d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.4d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_tensor_reduce_xor_tile_5d` | `nvvm.cp_async_bulk_tensor_reduce_xor_tile_5d` | LLVM `llvm.nvvm.cp.async.bulk.tensor.reduce.xor.tile.5d` from `int_nvvm_cp_async_bulk_tensor_reduce_xor_tile_5d` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5.3 Data Movement and Conversion Instructions: cp.reduce.async.bulk.tensor](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-reduce-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma-reduce`; expects `cp.reduce.async.bulk.tensor.5d.global.shared::cta.xor.tile.bulk_group [<address>], [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_wait_group` | `nvvm.cp_async_bulk_wait_group` | LLVM `llvm.nvvm.cp.async.bulk.wait.group` from `int_nvvm_cp_async_bulk_wait_group` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.wait_group <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::cp_async_bulk_wait_group_read` | `nvvm.cp_async_bulk_wait_group_read` | LLVM `llvm.nvvm.cp.async.bulk.wait.group.read` from `int_nvvm_cp_async_bulk_wait_group_read` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.9.26.5 Asynchronous bulk tensor copy](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-bulk-tensor)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `cp.async.bulk.wait_group.read <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_ca_16` | `nvvm.cp_async_ca_16` | LLVM `llvm.nvvm.cp.async.ca.shared.global.16` from `int_nvvm_cp_async_ca_shared_global_16` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.ca.shared.global [<address>], [<address>], 16;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_ca_4` | `nvvm.cp_async_ca_4` | LLVM `llvm.nvvm.cp.async.ca.shared.global.4` from `int_nvvm_cp_async_ca_shared_global_4` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.ca.shared.global [<address>], [<address>], 4;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_ca_8` | `nvvm.cp_async_ca_8` | LLVM `llvm.nvvm.cp.async.ca.shared.global.8` from `int_nvvm_cp_async_ca_shared_global_8` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.ca.shared.global [<address>], [<address>], 8;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_ca_zfill_16` | `nvvm.cp_async_ca_zfill_16` | LLVM `llvm.nvvm.cp.async.ca.shared.global.16.s` from `int_nvvm_cp_async_ca_shared_global_16_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.ca.shared.global [<address>], [<address>], 16, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_ca_zfill_4` | `nvvm.cp_async_ca_zfill_4` | LLVM `llvm.nvvm.cp.async.ca.shared.global.4.s` from `int_nvvm_cp_async_ca_shared_global_4_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.ca.shared.global [<address>], [<address>], 4, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_ca_zfill_8` | `nvvm.cp_async_ca_zfill_8` | LLVM `llvm.nvvm.cp.async.ca.shared.global.8.s` from `int_nvvm_cp_async_ca_shared_global_8_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.ca.shared.global [<address>], [<address>], 8, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_cg_16` | `nvvm.cp_async_cg_16` | LLVM `llvm.nvvm.cp.async.cg.shared.global.16` from `int_nvvm_cp_async_cg_shared_global_16` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.cg.shared.global [<address>], [<address>], 16;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_cg_zfill_16` | `nvvm.cp_async_cg_zfill_16` | LLVM `llvm.nvvm.cp.async.cg.shared.global.16.s` from `int_nvvm_cp_async_cg_shared_global_16_s` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.async](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.cg.shared.global [<address>], [<address>], 16, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_commit_group` | `nvvm.cp_async_commit_group` | LLVM `llvm.nvvm.cp.async.commit.group` from `int_nvvm_cp_async_commit_group` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.2 Data Movement and Conversion Instructions: cp.async.commit_group](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-commit-group)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.commit_group;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive` | `nvvm.cp_async_mbarrier_arrive` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive` from `int_nvvm_cp_async_mbarrier_arrive` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.mbarrier.arrive.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive_noinc` | `nvvm.cp_async_mbarrier_arrive_noinc` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive.noinc` from `int_nvvm_cp_async_mbarrier_arrive_noinc` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.mbarrier.arrive.noinc.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive_noinc_shared` | `nvvm.cp_async_mbarrier_arrive_noinc_shared` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive.noinc.shared` from `int_nvvm_cp_async_mbarrier_arrive_noinc_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.mbarrier.arrive.noinc.shared.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_mbarrier_arrive_shared` | `nvvm.cp_async_mbarrier_arrive_shared` | LLVM `llvm.nvvm.cp.async.mbarrier.arrive.shared` from `int_nvvm_cp_async_mbarrier_arrive_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.18 Parallel Synchronization and Communication Instructions: cp.async.mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.mbarrier.arrive.shared.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_wait_all` | `nvvm.cp_async_wait_all` | LLVM `llvm.nvvm.cp.async.wait.all` from `int_nvvm_cp_async_wait_all` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.3 Data Movement and Conversion Instructions: cp.async.wait_group / cp.async.wait_all](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.wait_all;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::async_copy::cp_async_wait_group` | `nvvm.cp_async_wait_group` | LLVM `llvm.nvvm.cp.async.wait.group` from `int_nvvm_cp_async_wait_group` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.26.3.3 Data Movement and Conversion Instructions: cp.async.wait_group / cp.async.wait_all](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cp.async.wait_group <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_f16x2_f32` | `nvvm.cvt_f16x2_f32` | LLVM `llvm.nvvm.ff2f16x2.rn` from `int_nvvm_ff2f16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.f16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_f32x2_bf16x2` | `nvvm.cvt_f32x2_bf16x2` | LLVM `llvm.nvvm.ff2bf16x2.rn` from `int_nvvm_ff2bf16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.bf16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_f16x2_e4m3x2` | `nvvm.cvt_rn_f16x2_e4m3x2` | LLVM `llvm.nvvm.e4m3x2.to.f16x2.rn` from `int_nvvm_e4m3x2_to_f16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.f16x2.e4m3x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_f16x2_e5m2x2` | `nvvm.cvt_rn_f16x2_e5m2x2` | LLVM `llvm.nvvm.e5m2x2.to.f16x2.rn` from `int_nvvm_e5m2x2_to_f16x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.f16x2.e5m2x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_relu_bf16x2_f32` | `nvvm.cvt_rn_relu_bf16x2_f32` | LLVM `llvm.nvvm.ff2bf16x2.rn.relu` from `int_nvvm_ff2bf16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.relu.bf16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_relu_f16x2_e4m3x2` | `nvvm.cvt_rn_relu_f16x2_e4m3x2` | LLVM `llvm.nvvm.e4m3x2.to.f16x2.rn.relu` from `int_nvvm_e4m3x2_to_f16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.relu.f16x2.e4m3x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_relu_f16x2_e5m2x2` | `nvvm.cvt_rn_relu_f16x2_e5m2x2` | LLVM `llvm.nvvm.e5m2x2.to.f16x2.rn.relu` from `int_nvvm_e5m2x2_to_f16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.relu.f16x2.e5m2x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_relu_f16x2_f32` | `nvvm.cvt_rn_relu_f16x2_f32` | LLVM `llvm.nvvm.ff2f16x2.rn.relu` from `int_nvvm_ff2f16x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.relu.f16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_relu_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn.relu.satfinite` from `int_nvvm_f2tf32_rn_relu_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.relu.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_relu_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn.relu` from `int_nvvm_f2tf32_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.relu.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_e4m3x2_f16x2` | `nvvm.cvt_rn_satfinite_e4m3x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e4m3x2.rn` from `int_nvvm_f16x2_to_e4m3x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.e4m3x2.f16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_e4m3x2_f32` | `nvvm.cvt_rn_satfinite_e4m3x2_f32` | LLVM `llvm.nvvm.ff.to.e4m3x2.rn` from `int_nvvm_ff_to_e4m3x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.e4m3x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_e5m2x2_f16x2` | `nvvm.cvt_rn_satfinite_e5m2x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e5m2x2.rn` from `int_nvvm_f16x2_to_e5m2x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.e5m2x2.f16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_e5m2x2_f32` | `nvvm.cvt_rn_satfinite_e5m2x2_f32` | LLVM `llvm.nvvm.ff.to.e5m2x2.rn` from `int_nvvm_ff_to_e5m2x2_rn` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.e5m2x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e4m3x2_f16x2` | `nvvm.cvt_rn_satfinite_relu_e4m3x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e4m3x2.rn.relu` from `int_nvvm_f16x2_to_e4m3x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.relu.e4m3x2.f16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e4m3x2_f32` | `nvvm.cvt_rn_satfinite_relu_e4m3x2_f32` | LLVM `llvm.nvvm.ff.to.e4m3x2.rn.relu` from `int_nvvm_ff_to_e4m3x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.relu.e4m3x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e5m2x2_f16x2` | `nvvm.cvt_rn_satfinite_relu_e5m2x2_f16x2` | LLVM `llvm.nvvm.f16x2.to.e5m2x2.rn.relu` from `int_nvvm_f16x2_to_e5m2x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.relu.e5m2x2.f16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_relu_e5m2x2_f32` | `nvvm.cvt_rn_satfinite_relu_e5m2x2_f32` | LLVM `llvm.nvvm.ff.to.e5m2x2.rn.relu` from `int_nvvm_ff_to_e5m2x2_rn_relu` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_89+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.relu.e5m2x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn.satfinite` from `int_nvvm_f2tf32_rn_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rn_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rn` from `int_nvvm_f2tf32_rn` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rn.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rna_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rna.satfinite` from `int_nvvm_f2tf32_rna_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.1 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rna.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rna_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rna` from `int_nvvm_f2tf32_rna` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rna.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rz_bf16x2_f32` | `nvvm.cvt_rz_bf16x2_f32` | LLVM `llvm.nvvm.ff2bf16x2.rz` from `int_nvvm_ff2bf16x2_rz` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rz.bf16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rz_f16x2_f32` | `nvvm.cvt_rz_f16x2_f32` | LLVM `llvm.nvvm.ff2f16x2.rz` from `int_nvvm_ff2f16x2_rz` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rz.f16x2.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rz_relu_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz.relu.satfinite` from `int_nvvm_f2tf32_rz_relu_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rz.relu.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rz_relu_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz.relu` from `int_nvvm_f2tf32_rz_relu` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rz.relu.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rz_satfinite_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz.satfinite` from `int_nvvm_f2tf32_rz_satfinite` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rz.satfinite.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::convert::cvt_rz_tf32_f32` | `nvvm.scalar_conversion` | LLVM `llvm.nvvm.f2tf32.rz` from `int_nvvm_f2tf32_rz` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.22 Data Movement and Conversion Instructions: cvt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `cvt.rz.tf32.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rm.f` from `int_nvvm_div_rm_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rm.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rm.d` from `int_nvvm_div_rm_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rm.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rm.ftz.f` from `int_nvvm_div_rm_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rm.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rn.f` from `int_nvvm_div_rn_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rn.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rn.d` from `int_nvvm_div_rn_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rn.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rn.ftz.f` from `int_nvvm_div_rn_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rn.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rp.f` from `int_nvvm_div_rp_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rp.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rp.d` from `int_nvvm_div_rp_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rp.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rp.ftz.f` from `int_nvvm_div_rp_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rp.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rz.f` from `int_nvvm_div_rz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rz.d` from `int_nvvm_div_rz_d` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rz.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::div_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.div.rz.ftz.f` from `int_nvvm_div_rz_ftz_f` | safe; scope thread; impure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.8 Floating Point Instructions: div](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-div)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `div.rz.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::dotprod::dp2a_s32` | `nvvm.dp2a_s32` | LLVM `llvm.nvvm.idp2a.s.s` from `int_nvvm_idp2a_s_s` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.25 Integer Arithmetic Instructions: dp2a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `dp2a.lo.s32.s32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::dotprod::dp2a_u32` | `nvvm.dp2a_u32` | LLVM `llvm.nvvm.idp2a.u.u` from `int_nvvm_idp2a_u_u` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.25 Integer Arithmetic Instructions: dp2a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `dp2a.lo.u32.u32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::dotprod::dp4a_s32` | `nvvm.dp4a_s32` | LLVM `llvm.nvvm.idp4a.s.s` from `int_nvvm_idp4a_s_s` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.24 Integer Arithmetic Instructions: dp4a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `dp4a.s32.s32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::dotprod::dp4a_u32` | `nvvm.dp4a_u32` | LLVM `llvm.nvvm.idp4a.u.u` from `int_nvvm_idp4a_u_u` | safe; scope thread; pure; memory none; convergent false | PTX 5.0 on sm_61+ ([PTX ISA 9.3, 9.7.1.24 Integer Arithmetic Instructions: dp4a](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `dp4a.u32.u32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::dsmem_read_u32` | `nvvm.dsmem_read_u32` | PTX-native `mapa.shared::cluster.u64 + ld.shared::cluster.u32` | unsafe; scope cluster; impure; memory read; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.8 Data Movement and Conversion Instructions: ld](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-ld)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ld.shared::cluster.u32 <register>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::shared::dynamic_smem_size` | `nvvm.read_ptx_sreg_dynamic_smem_size` | LLVM `llvm.nvvm.read.ptx.sreg.dynamic_smem_size` from `int_nvvm_read_ptx_sreg_dynamic_smem_size` | safe; scope cta; pure; memory none; convergent false | PTX 4.1 on sm_20+ ([PTX ISA 9.3, 10.32 Special Registers: %dynamic_smem_size](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-dynamic-smem-size)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %dynamic_smem_size;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::elect_sync` | `nvvm.elect_sync` | LLVM `llvm.nvvm.elect.sync` from `int_nvvm_elect_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.14.15 Parallel Synchronization and Communication Instructions: elect.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-elect-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `elect.sync <register\|predicate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::grid::envreg1` | `nvvm.read_ptx_sreg_envreg1` | LLVM `llvm.nvvm.read.ptx.sreg.envreg1` from `int_nvvm_read_ptx_sreg_envreg1` | safe; scope grid; pure; memory none; convergent false | PTX 2.1 on all ([PTX ISA 9.3, 10.27 Special Registers: %envreg<32>](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-envreg)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.b32 <register>, %envreg1;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::grid::envreg2` | `nvvm.read_ptx_sreg_envreg2` | LLVM `llvm.nvvm.read.ptx.sreg.envreg2` from `int_nvvm_read_ptx_sreg_envreg2` | safe; scope grid; pure; memory none; convergent false | PTX 2.1 on all ([PTX ISA 9.3, 10.27 Special Registers: %envreg<32>](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-envreg)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.b32 <register>, %envreg2;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::ex2_approx_f16` | `nvvm.scalar_math` | LLVM `llvm.nvvm.ex2.approx` from `int_nvvm_ex2_approx` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_75+ ([PTX ISA 9.3, 9.7.4.10 Half Precision Floating Point Instructions: ex2](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-ex2)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ex2.approx.f16 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::ex2_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.ex2.approx` from `int_nvvm_ex2_approx` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.13 Floating Point Instructions: ex2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-ex2)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ex2.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::ex2_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.ex2.approx.ftz` from `int_nvvm_ex2_approx_ftz` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.13 Floating Point Instructions: ex2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-ex2)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ex2.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::fence_mbarrier_init_release_cluster` | `nvvm.fence_mbarrier_init_release_cluster` | LLVM `llvm.nvvm.fence.mbarrier_init.release.cluster` from `int_nvvm_fence_mbarrier_init_release_cluster` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.mbarrier_init](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-mbarrier-init)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fence.mbarrier_init.release.cluster;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::fence_proxy_async_generic_acquire_shared_cluster_cluster` | `nvvm.fence_proxy_async_generic_acquire_shared_cluster_cluster` | LLVM `llvm.nvvm.fence.proxy.async_generic.acquire.sync_restrict.space.cluster.scope.cluster` from `int_nvvm_fence_proxy_async_generic_acquire_sync_restrict_space_cluster_scope_cluster` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.6 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.proxy](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fence.proxy.async::generic.acquire.sync_restrict::shared::cluster.cluster;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::fence_proxy_async_generic_release_shared_cta_cluster` | `nvvm.fence_proxy_async_generic_release_shared_cta_cluster` | LLVM `llvm.nvvm.fence.proxy.async_generic.release.sync_restrict.space.cta.scope.cluster` from `int_nvvm_fence_proxy_async_generic_release_sync_restrict_space_cta_scope_cluster` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.6 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.proxy](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fence.proxy.async::generic.release.sync_restrict::shared::cta.cluster;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::fence_proxy_async_shared_cta` | `nvvm.fence_proxy_async_shared_cta` | LLVM `llvm.nvvm.fence.proxy.async.shared_cta` from `int_nvvm_fence_proxy_async_shared_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: fence.proxy](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-fence-proxy)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fence.proxy.async.shared::cta;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_cluster` | `nvvm.fence_proxy_tensormap_generic_acquire_cluster` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.cluster` from `int_nvvm_fence_proxy_tensormap_generic_acquire_cluster` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.acquire.cluster [<address>], 128;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_cta` | `nvvm.fence_proxy_tensormap_generic_acquire_cta` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.cta` from `int_nvvm_fence_proxy_tensormap_generic_acquire_cta` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.acquire.cta [<address>], 128;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_gpu` | `nvvm.fence_proxy_tensormap_generic_acquire_gpu` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.gpu` from `int_nvvm_fence_proxy_tensormap_generic_acquire_gpu` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.acquire.gpu [<address>], 128;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_acquire_system` | `nvvm.fence_proxy_tensormap_generic_acquire_system` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.acquire.sys` from `int_nvvm_fence_proxy_tensormap_generic_acquire_sys` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.acquire.sys [<address>], 128;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_cluster` | `nvvm.fence_proxy_tensormap_generic_release_cluster` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.cluster` from `int_nvvm_fence_proxy_tensormap_generic_release_cluster` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.release.cluster;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_cta` | `nvvm.fence_proxy_tensormap_generic_release_cta` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.cta` from `int_nvvm_fence_proxy_tensormap_generic_release_cta` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.release.cta;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_gpu` | `nvvm.fence_proxy_tensormap_generic_release_gpu` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.gpu` from `int_nvvm_fence_proxy_tensormap_generic_release_gpu` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.release.gpu;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::fence_proxy_tensormap_generic_release_system` | `nvvm.fence_proxy_tensormap_generic_release_system` | LLVM `llvm.nvvm.fence.proxy.tensormap_generic.release.sys` from `int_nvvm_fence_proxy_tensormap_generic_release_sys` | safe; scope thread; impure; memory read_write; convergent false | PTX 8.3 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `fence.proxy.tensormap::generic.release.sys;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::fma_bf16x2` | `nvvm.fma_bf16x2` | LLVM `llvm.nvvm.fma.rn.bf16x2` from `int_nvvm_fma_rn_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.bf16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::fma_f16x2` | `nvvm.fma_f16x2` | LLVM `llvm.nvvm.fma.rn.f16x2` from `int_nvvm_fma_rn_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::fma_f32x2` | `nvvm.fma_f32x2` | PTX-native `fma.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.4 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.f32x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::fma_ftz_f16x2` | `nvvm.fma_ftz_f16x2` | LLVM `llvm.nvvm.fma.rn.ftz.f16x2` from `int_nvvm_fma_rn_ftz_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.ftz.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::fma_ftz_f32x2` | `nvvm.fma_ftz_f32x2` | PTX-native `fma.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.4 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.ftz.f32x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::fma_ftz_relu_f16x2` | `nvvm.fma_ftz_relu_f16x2` | LLVM `llvm.nvvm.fma.rn.ftz.relu.f16x2` from `int_nvvm_fma_rn_ftz_relu_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.ftz.relu.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::fma_ftz_sat_f16x2` | `nvvm.fma_ftz_sat_f16x2` | LLVM `llvm.nvvm.fma.rn.ftz.sat.f16x2` from `int_nvvm_fma_rn_ftz_sat_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.ftz.sat.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::fma_relu_bf16x2` | `nvvm.fma_relu_bf16x2` | LLVM `llvm.nvvm.fma.rn.relu.bf16x2` from `int_nvvm_fma_rn_relu_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.relu.bf16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::fma_relu_f16x2` | `nvvm.fma_relu_f16x2` | LLVM `llvm.nvvm.fma.rn.relu.f16x2` from `int_nvvm_fma_rn_relu_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.relu.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.f` from `int_nvvm_fma_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rm.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.d` from `int_nvvm_fma_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rm.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.ftz.f` from `int_nvvm_fma_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rm.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rm_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.ftz.sat.f` from `int_nvvm_fma_rm_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rm.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rm_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rm.sat.f` from `int_nvvm_fma_rm_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rm.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.f` from `int_nvvm_fma_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.d` from `int_nvvm_fma_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.ftz.f` from `int_nvvm_fma_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rn_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.ftz.sat.f` from `int_nvvm_fma_rn_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rn_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rn.sat.f` from `int_nvvm_fma_rn_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.f` from `int_nvvm_fma_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rp.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.d` from `int_nvvm_fma_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rp.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.ftz.f` from `int_nvvm_fma_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rp.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rp_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.ftz.sat.f` from `int_nvvm_fma_rp_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rp.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rp_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rp.sat.f` from `int_nvvm_fma_rp_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rp.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.f` from `int_nvvm_fma_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.d` from `int_nvvm_fma_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rz.f64 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.ftz.f` from `int_nvvm_fma_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rz.ftz.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rz_ftz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.ftz.sat.f` from `int_nvvm_fma_rz_ftz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rz.ftz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::fma_rz_sat_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.fma.rz.sat.f` from `int_nvvm_fma_rz_sat_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.6 Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rz.sat.f32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::fma_sat_f16x2` | `nvvm.fma_sat_f16x2` | LLVM `llvm.nvvm.fma.rn.sat.f16x2` from `int_nvvm_fma_rn_sat_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.4 Half Precision Floating Point Instructions: fma](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `fma.rn.sat.f16x2 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::debug::globaltimer` | `nvvm.read_ptx_sreg_globaltimer` | LLVM `llvm.nvvm.read.ptx.sreg.globaltimer` from `int_nvvm_read_ptx_sreg_globaltimer` | safe; scope device; impure; memory inaccessible_read_write; convergent false | PTX 3.1 on sm_30+ ([PTX ISA 9.3, 10.28 Special Registers: %globaltimer, %globaltimer_lo, %globaltimer_hi](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-globaltimer)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u64 <register>, %globaltimer;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::grid::grid_dependency_launch_dependents` | `nvvm.grid_dependency_launch_dependents` | LLVM `llvm.nvvm.griddepcontrol.launch.dependents` from `int_nvvm_griddepcontrol_launch_dependents` | unsafe; scope grid; impure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.15.1 Parallel Synchronization and Communication Instructions: griddepcontrol](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `griddepcontrol.launch_dependents;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::grid::grid_dependency_wait` | `nvvm.grid_dependency_wait` | LLVM `llvm.nvvm.griddepcontrol.wait` from `int_nvvm_griddepcontrol_wait` | unsafe; scope grid; impure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.15.1 Parallel Synchronization and Communication Instructions: griddepcontrol](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `griddepcontrol.wait;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::grid_dim_x` | `nvvm.read_ptx_sreg_nctaid_x` | LLVM `llvm.nvvm.read.ptx.sreg.nctaid.x` from `int_nvvm_read_ptx_sreg_nctaid_x` | safe; scope grid; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.7 Special Registers: %nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nctaid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::grid_dim_y` | `nvvm.read_ptx_sreg_nctaid_y` | LLVM `llvm.nvvm.read.ptx.sreg.nctaid.y` from `int_nvvm_read_ptx_sreg_nctaid_y` | safe; scope grid; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.7 Special Registers: %nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nctaid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::grid_dim_z` | `nvvm.read_ptx_sreg_nctaid_z` | LLVM `llvm.nvvm.read.ptx.sreg.nctaid.z` from `int_nvvm_read_ptx_sreg_nctaid_z` | safe; scope grid; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.7 Special Registers: %nctaid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nctaid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nctaid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::thread::gridid` | `nvvm.read_ptx_sreg_gridid` | PTX-native `mov.u64 %gridid` | safe; scope grid; pure; memory none; convergent false | PTX 3.0 on sm_30+ ([PTX ISA 9.3, 10.10 Special Registers: %gridid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-gridid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u64 <register>, %gridid;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::lane_id` | `nvvm.read_ptx_sreg_laneid` | LLVM `llvm.nvvm.read.ptx.sreg.laneid` from `int_nvvm_read_ptx_sreg_laneid` | safe; scope thread; pure; memory none; convergent false | PTX 1.3 on all ([PTX ISA 9.3, 10.3 Special Registers: %laneid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-laneid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %laneid;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::lanemask_eq` | `nvvm.read_ptx_sreg_lanemask_eq` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.eq` from `int_nvvm_read_ptx_sreg_lanemask_eq` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.11 Special Registers: %lanemask_eq](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-eq)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %lanemask_eq;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::lanemask_ge` | `nvvm.read_ptx_sreg_lanemask_ge` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.ge` from `int_nvvm_read_ptx_sreg_lanemask_ge` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.14 Special Registers: %lanemask_ge](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-ge)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %lanemask_ge;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::lanemask_gt` | `nvvm.read_ptx_sreg_lanemask_gt` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.gt` from `int_nvvm_read_ptx_sreg_lanemask_gt` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.15 Special Registers: %lanemask_gt](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-gt)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %lanemask_gt;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::lanemask_le` | `nvvm.read_ptx_sreg_lanemask_le` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.le` from `int_nvvm_read_ptx_sreg_lanemask_le` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.12 Special Registers: %lanemask_le](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-le)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %lanemask_le;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::lanemask_lt` | `nvvm.read_ptx_sreg_lanemask_lt` | LLVM `llvm.nvvm.read.ptx.sreg.lanemask.lt` from `int_nvvm_read_ptx_sreg_lanemask_lt` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.13 Special Registers: %lanemask_lt](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-lanemask-lt)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %lanemask_lt;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m16n16_x1_trans_b8` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x1.trans.b8` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x1_trans_b8` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m16n16.x1.trans.shared.b8 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x1.trans.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x1_trans_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m16n16.x1.trans.shared.b8x16.b4x16_p64 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x1.trans.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x1_trans_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m16n16.x1.trans.shared.b8x16.b6x16_p32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m16n16_x2_trans_b8` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x2.trans.b8` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x2_trans_b8` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m16n16.x2.trans.shared.b8 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x2.trans.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x2_trans_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m16n16.x2.trans.shared.b8x16.b4x16_p64 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m16n16.x2.trans.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m16n16_x2_trans_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m16n16.x2.trans.shared.b8x16.b6x16_p32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n16_x1_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x1.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x1_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n16.x1.shared.b8x16.b4x16_p64 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n16_x1_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x1.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x1_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n16.x1.shared.b8x16.b6x16_p32 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n16_x2_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x2.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x2_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n16.x2.shared.b8x16.b4x16_p64 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n16_x2_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x2.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x2_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n16.x2.shared.b8x16.b6x16_p32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n16_x4_b8x16_b4x16_p64` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x4.b8x16.b4x16_p64` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x4_b8x16_b4x16_p64` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n16.x4.shared.b8x16.b4x16_p64 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n16_x4_b8x16_b6x16_p32` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n16.x4.b8x16.b6x16_p32` from `int_nvvm_ldmatrix_sync_aligned_m8n16_x4_b8x16_b6x16_p32` | unsafe; scope warp; impure; memory read; convergent true | Instruction floor PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f; effective LLVM target floors: sm_100a PTX 8.6, sm_100f PTX 8.8, sm_103a PTX 8.8, sm_103f PTX 8.8, sm_110a PTX 9.0, sm_110f PTX 9.0, sm_120a PTX 8.7, sm_120f PTX 8.8, sm_121a PTX 8.8, sm_121f PTX 8.8 ([PTX ISA 9.3, 9.7.15.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n16.x4.shared.b8x16.b6x16_p32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n8_x1_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x1_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n8.x1.shared.b16 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n8_x1_trans_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x1.trans.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x1_trans_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n8.x1.trans.shared.b16 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n8_x2_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x2_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n8.x2.shared.b16 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n8_x2_trans_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x2.trans.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x2_trans_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n8_x4_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x4_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n8.x4.shared.b16 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::ldmatrix_m8n8_x4_trans_b16` | `nvvm.ldmatrix` | LLVM `llvm.nvvm.ldmatrix.sync.aligned.m8n8.x4.trans.b16` from `int_nvvm_ldmatrix_sync_aligned_m8n8_x4_trans_b16` | unsafe; scope warp; impure; memory read; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.14.5.15 Warp-level matrix load instruction: ldmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::lg2_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.lg2.approx.f` from `int_nvvm_lg2_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.12 Floating Point Instructions: lg2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-lg2)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `lg2.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::lg2_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.lg2.approx.ftz.f` from `int_nvvm_lg2_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.12 Floating Point Instructions: lg2](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-lg2)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `lg2.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::cluster::map_shared_rank` | `nvvm.mapa_shared_cluster` | LLVM `llvm.nvvm.mapa.shared.cluster` from `int_nvvm_mapa_shared_cluster` | unsafe; scope cluster; impure; memory none; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.9.24 Data Movement and Conversion Instructions: mapa](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-mapa)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mapa.shared::cluster.u64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::match_all_i64_sync` | `nvvm.match_all_sync_i64` | LLVM `llvm.nvvm.match.all.sync.i64p` from `int_nvvm_match_all_sync_i64p` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `match.all.sync.b64 <register\|predicate>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::match_all_sync` | `nvvm.match_all_sync_i32` | LLVM `llvm.nvvm.match.all.sync.i32p` from `int_nvvm_match_all_sync_i32p` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `match.all.sync.b32 <register\|predicate>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::match_any_i64_sync` | `nvvm.match_any_sync_i64` | LLVM `llvm.nvvm.match.any.sync.i64` from `int_nvvm_match_any_sync_i64` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `match.any.sync.b64 <register>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::match_any_sync` | `nvvm.match_any_sync_i32` | LLVM `llvm.nvvm.match.any.sync.i32` from `int_nvvm_match_any_sync_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.14.11 Parallel Synchronization and Communication Instructions: match.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-match-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `match.any.sync.b32 <register>, <register-or-immediate>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::max_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.bf16` from `int_nvvm_fmax_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::max_bf16x2` | `nvvm.max_bf16x2` | LLVM `llvm.nvvm.fmax.bf16x2` from `int_nvvm_fmax_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.f16` from `int_nvvm_fmax_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_f16x2` | `nvvm.max_f16x2` | LLVM `llvm.nvvm.fmax.f16x2` from `int_nvvm_fmax_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_ftz_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.f16` from `int_nvvm_fmax_ftz_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_ftz_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.f16x2` from `int_nvvm_fmax_ftz_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_ftz_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.f16` from `int_nvvm_fmax_ftz_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_ftz_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.f16x2` from `int_nvvm_fmax_ftz_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_ftz_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16` from `int_nvvm_fmax_ftz_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_ftz_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.xorsign.abs.f16x2` from `int_nvvm_fmax_ftz_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::max_ftz_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.nan.xorsign.abs.f` from `int_nvvm_fmax_ftz_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_ftz_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.xorsign.abs.f16` from `int_nvvm_fmax_ftz_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_ftz_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.xorsign.abs.f16x2` from `int_nvvm_fmax_ftz_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::max_ftz_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.ftz.xorsign.abs.f` from `int_nvvm_fmax_ftz_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.ftz.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::max_nan_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.bf16` from `int_nvvm_fmax_nan_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::max_nan_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.bf16x2` from `int_nvvm_fmax_nan_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.f16` from `int_nvvm_fmax_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.f16x2` from `int_nvvm_fmax_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::max_nan_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.bf16` from `int_nvvm_fmax_nan_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::max_nan_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.bf16x2` from `int_nvvm_fmax_nan_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.f16` from `int_nvvm_fmax_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.f16x2` from `int_nvvm_fmax_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::max_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.nan.xorsign.abs.f` from `int_nvvm_fmax_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::i16x2::max_relu_s16x2` | `nvvm.max_relu_s16x2` | PTX-native `max.relu.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `max.relu.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::int::max_relu_s32` | `nvvm.max_relu_s32` | PTX-native `max.relu.s32` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `max.relu.s32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::i16x2::max_s16x2` | `nvvm.max_s16x2` | PTX-native `max.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `max.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::i16x2::max_u16x2` | `nvvm.max_u16x2` | PTX-native `max.u16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.14 Integer Arithmetic Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `max.u16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::max_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.bf16` from `int_nvvm_fmax_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::max_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.bf16x2` from `int_nvvm_fmax_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::max_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.f16` from `int_nvvm_fmax_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::max_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.f16x2` from `int_nvvm_fmax_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.8 Half Precision Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::max_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmax.xorsign.abs.f` from `int_nvvm_fmax_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: max](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-max)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `max.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_arrive` | `nvvm.mbarrier_arrive_shared` | LLVM `llvm.nvvm.mbarrier.arrive.shared` from `int_nvvm_mbarrier_arrive_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.16 Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.arrive.shared.b64 <register>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_arrive_cluster` | `nvvm.mbarrier_arrive_cluster` | PTX-native `mbarrier.arrive.release.cluster.shared::cluster.b64` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.arrive.release.cluster.shared::cluster.b64 _, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_arrive_expect_tx` | `nvvm.mbarrier_arrive_expect_tx_shared` | LLVM `llvm.nvvm.mbarrier.arrive.expect.tx.scope.cta.space.cta` from `int_nvvm_mbarrier_arrive_expect_tx_scope_cta_space_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.arrive.expect_tx.release.cta.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_arrive_expect_tx_cluster` | `nvvm.mbarrier_arrive_expect_tx_cluster` | LLVM `llvm.nvvm.mbarrier.arrive.expect.tx.relaxed.scope.cluster.space.cta` from `int_nvvm_mbarrier_arrive_expect_tx_relaxed_scope_cluster_space_cta` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.6 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.arrive.expect_tx.relaxed.cluster.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_arrive_no_complete` | `nvvm.mbarrier_arrive_no_complete_shared` | LLVM `llvm.nvvm.mbarrier.arrive.noComplete.shared` from `int_nvvm_mbarrier_arrive_noComplete_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.16 Parallel Synchronization and Communication Instructions: mbarrier.arrive](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.arrive.noComplete.shared.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_init` | `nvvm.mbarrier_init_shared` | LLVM `llvm.nvvm.mbarrier.init.shared` from `int_nvvm_mbarrier_init_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.12 Parallel Synchronization and Communication Instructions: mbarrier.init](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-init)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.init.shared.b64 [<address>], <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_inval` | `nvvm.mbarrier_inval_shared` | LLVM `llvm.nvvm.mbarrier.inval.shared` from `int_nvvm_mbarrier_inval_shared` | unsafe; scope cta; impure; memory write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.13 Parallel Synchronization and Communication Instructions: mbarrier.inval](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-inval)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.inval.shared.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_test_wait` | `nvvm.mbarrier_test_wait_shared` | LLVM `llvm.nvvm.mbarrier.test.wait.shared` from `int_nvvm_mbarrier_test_wait_shared` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.16.19 Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.test_wait.shared.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_try_wait` | `nvvm.mbarrier_try_wait_shared` | LLVM `llvm.nvvm.mbarrier.try.wait.scope.cta.space.cta` from `int_nvvm_mbarrier_try_wait_scope_cta_space_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.try_wait.shared.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_try_wait_parity` | `nvvm.mbarrier_try_wait_parity_shared` | LLVM `llvm.nvvm.mbarrier.try.wait.parity.scope.cta.space.cta` from `int_nvvm_mbarrier_try_wait_parity_scope_cta_space_cta` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.try_wait.parity.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::mbarrier_try_wait_parity_cluster` | `nvvm.mbarrier_try_wait_parity_cluster` | LLVM `llvm.nvvm.mbarrier.try.wait.parity.scope.cluster.space.cta` from `int_nvvm_mbarrier_try_wait_parity_scope_cluster_space_cta` | unsafe; scope cluster; impure; memory read_write; convergent true | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: mbarrier.test_wait / mbarrier.try_wait](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mbarrier.try_wait.parity.acquire.cluster.shared::cta.b64 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::min_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.bf16` from `int_nvvm_fmin_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::min_bf16x2` | `nvvm.min_bf16x2` | LLVM `llvm.nvvm.fmin.bf16x2` from `int_nvvm_fmin_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.f16` from `int_nvvm_fmin_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_f16x2` | `nvvm.min_f16x2` | LLVM `llvm.nvvm.fmin.f16x2` from `int_nvvm_fmin_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_ftz_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.f16` from `int_nvvm_fmin_ftz_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_ftz_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.f16x2` from `int_nvvm_fmin_ftz_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_ftz_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.f16` from `int_nvvm_fmin_ftz_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_ftz_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.f16x2` from `int_nvvm_fmin_ftz_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_ftz_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16` from `int_nvvm_fmin_ftz_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_ftz_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.xorsign.abs.f16x2` from `int_nvvm_fmin_ftz_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::min_ftz_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.nan.xorsign.abs.f` from `int_nvvm_fmin_ftz_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_ftz_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.xorsign.abs.f16` from `int_nvvm_fmin_ftz_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_ftz_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.xorsign.abs.f16x2` from `int_nvvm_fmin_ftz_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::min_ftz_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.ftz.xorsign.abs.f` from `int_nvvm_fmin_ftz_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.ftz.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::min_nan_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.bf16` from `int_nvvm_fmin_nan_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::min_nan_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.bf16x2` from `int_nvvm_fmin_nan_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_nan_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.f16` from `int_nvvm_fmin_nan_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_nan_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.f16x2` from `int_nvvm_fmin_nan_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::min_nan_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.bf16` from `int_nvvm_fmin_nan_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::min_nan_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.bf16x2` from `int_nvvm_fmin_nan_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_nan_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.f16` from `int_nvvm_fmin_nan_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_nan_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.f16x2` from `int_nvvm_fmin_nan_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::min_nan_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.nan.xorsign.abs.f` from `int_nvvm_fmin_nan_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.NaN.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::i16x2::min_relu_s16x2` | `nvvm.min_relu_s16x2` | PTX-native `min.relu.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `min.relu.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::int::min_relu_s32` | `nvvm.min_relu_s32` | PTX-native `min.relu.s32` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `min.relu.s32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::i16x2::min_s16x2` | `nvvm.min_s16x2` | PTX-native `min.s16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `min.s16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::i16x2::min_u16x2` | `nvvm.min_u16x2` | PTX-native `min.u16x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, 9.7.1.13 Integer Arithmetic Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `min.u16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16::min_xorsign_abs_bf16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.bf16` from `int_nvvm_fmin_xorsign_abs_bf16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.xorsign.abs.bf16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::min_xorsign_abs_bf16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.bf16x2` from `int_nvvm_fmin_xorsign_abs_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.xorsign.abs.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16::min_xorsign_abs_f16` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.f16` from `int_nvvm_fmin_xorsign_abs_f16` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.xorsign.abs.f16 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::min_xorsign_abs_f16x2` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.f16x2` from `int_nvvm_fmin_xorsign_abs_f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.4.7 Half Precision Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.xorsign.abs.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::min_xorsign_abs_f32` | `nvvm.extended_minmax` | LLVM `llvm.nvvm.fmin.xorsign.abs.f` from `int_nvvm_fmin_xorsign_abs_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.2 on sm_86+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: min](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-min)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `min.xorsign.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k128_s32_b1_and_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.and.popc.m16n8k128.row.col.b1` from `int_nvvm_mma_and_popc_m16n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k128.row.col.s32.b1.b1.s32.and.popc <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k128_s32_b1_xor_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.xor.popc.m16n8k128.row.col.b1` from `int_nvvm_mma_xor_popc_m16n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k128.row.col.s32.b1.b1.s32.xor.popc <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_f16_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_f32_bf16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.bf16` from `int_nvvm_mma_m16n8k16_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_f32_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e4m3.e4m3.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e4m3.e5m2.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e5m2.e4m3.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f16_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f16.e5m2.e5m2.f16` from `int_nvvm_mma_m16n8k16_row_col_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e4m3.e4m3.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e4m3.e5m2.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e5m2.e4m3.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_fp8_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.f32.e5m2.e5m2.f32` from `int_nvvm_mma_m16n8k16_row_col_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.s8` from `int_nvvm_mma_m16n8k16_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.s8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.s8.u8` from `int_nvvm_mma_m16n8k16_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_s8_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.s8.u8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.u8` from `int_nvvm_mma_m16n8k16_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.u8.s8` from `int_nvvm_mma_m16n8k16_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.u8.s8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k16_s32_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k16.row.col.satfinite.u8` from `int_nvvm_mma_m16n8k16_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k16.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k256_s32_b1_and_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.and.popc.m16n8k256.row.col.b1` from `int_nvvm_mma_and_popc_m16n8k256_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k256.row.col.s32.b1.b1.s32.and.popc <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k256_s32_b1_xor_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.xor.popc.m16n8k256.row.col.b1` from `int_nvvm_mma_xor_popc_m16n8k256_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k256.row.col.s32.b1.b1.s32.xor.popc <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m1_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m1_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m1.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e2m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e2m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e2m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e3m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e3m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e3m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f16_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m1_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m1_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m1.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e2m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e2m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e2m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e3m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e3m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e3m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_kind_f8f6f4_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::f8f6f4.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e4m3.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e4m3.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e5m2.e4m3.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f16_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f16.e5m2.e5m2.f16` from `int_nvvm_mma_m16n8k32_row_col_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:2>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e4m3.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e4m3.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e5m2.e4m3.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_fp8_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.f32.e5m2.e5m2.f32` from `int_nvvm_mma_m16n8k32_row_col_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m1.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m1_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m1.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e2m3.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e2m3_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e2m3.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e3m2.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e3m2_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e3m2.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e4m3.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e4m3_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e4m3.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e2m1.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e2m1_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e2m1.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e2m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e2m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e2m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e3m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e3m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e3m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e4m3.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e4m3_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e4m3.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.block.scale.m16n8k32.row.col.mxf8f6f4.f32.e5m2.e5m2.f32.ue8m0` from `int_nvvm_mma_block_scale_m16n8k32_row_col_mxf8f6f4_f32_e5m2_e5m2_f32_ue8m0` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.kind::mxf8f6f4.block_scale.f32.e5m2.e5m2.f32.ue8m0 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>, <register>, <register-list:2>, <register>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s4` from `int_nvvm_mma_m16n8k32_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s4.u4` from `int_nvvm_mma_m16n8k32_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s4_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s4.u4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s8` from `int_nvvm_mma_m16n8k32_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.s8.u8` from `int_nvvm_mma_m16n8k32_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_s8_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.s8.u8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u4` from `int_nvvm_mma_m16n8k32_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u4.s4` from `int_nvvm_mma_m16n8k32_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u4.s4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u4` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u8` from `int_nvvm_mma_m16n8k32_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.u8.s8` from `int_nvvm_mma_m16n8k32_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u8.s8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k32_s32_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k32.row.col.satfinite.u8` from `int_nvvm_mma_m16n8k32_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k4_f32_tf32` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k4.row.col.tf32` from `int_nvvm_mma_m16n8k4_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k4.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.s4` from `int_nvvm_mma_m16n8k64_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.s4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.s4.u4` from `int_nvvm_mma_m16n8k64_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_s4_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.s4.u4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.u4` from `int_nvvm_mma_m16n8k64_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.u4.s4` from `int_nvvm_mma_m16n8k64_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.u4.s4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k64_s32_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k64.row.col.satfinite.u4` from `int_nvvm_mma_m16n8k64_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k8_f16_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.f16.f16` from `int_nvvm_mma_m16n8k8_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k8.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:2>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k8_f32_bf16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.bf16` from `int_nvvm_mma_m16n8k8_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k8.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k8_f32_f16` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.f32.f32` from `int_nvvm_mma_m16n8k8_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:2>, <register-list:1>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m16n8k8_f32_tf32` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m16n8k8.row.col.tf32` from `int_nvvm_mma_m16n8k8_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:4>, <register-list:2>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k128_s32_b1_and_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.and.popc.m8n8k128.row.col.b1` from `int_nvvm_mma_and_popc_m8n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k128.row.col.s32.b1.b1.s32.and.popc <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k128_s32_b1_xor_popc` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.xor.popc.m8n8k128.row.col.b1` from `int_nvvm_mma_xor_popc_m8n8k128_row_col_b1` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k128.row.col.s32.b1.b1.s32.xor.popc <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.s8` from `int_nvvm_mma_m8n8k16_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.s32.s8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.s8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.s8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.s8.u8` from `int_nvvm_mma_m8n8k16_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.s32.s8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_s8_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.s8.u8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.s8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.u8` from `int_nvvm_mma_m8n8k16_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.s32.u8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8_s8` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.u8.s8` from `int_nvvm_mma_m8n8k16_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.s32.u8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8_s8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.u8.s8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.u8.s8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k16_s32_u8_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k16.row.col.satfinite.u8` from `int_nvvm_mma_m8n8k16_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k16.row.col.satfinite.s32.u8.u8.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.s4` from `int_nvvm_mma_m8n8k32_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.s32.s4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.s4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.s4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.s4.u4` from `int_nvvm_mma_m8n8k32_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.s32.s4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_s4_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.s4.u4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.s4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.u4` from `int_nvvm_mma_m8n8k32_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.s32.u4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4_s4` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.u4.s4` from `int_nvvm_mma_m8n8k32_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.s32.u4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4_s4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.u4.s4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.u4.s4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k32_s32_u4_satfinite` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k32.row.col.satfinite.u4` from `int_nvvm_mma_m8n8k32_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 6.5 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k32.row.col.satfinite.s32.u4.u4.s32 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_m8n8k4_f64` | `nvvm.register_mma` | LLVM `llvm.nvvm.mma.m8n8k4.row.col.f64` from `int_nvvm_mma_m8n8k4_row_col_f64` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.15.5.14 Multiply-and-Accumulate Instruction: mma](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sync.aligned.m8n8k4.row.col.f64.f64.f64.f64 <register-list:2>, <register-list:1>, <register-list:1>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.s4.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.u4.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k128_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k128.row.col.satfinite.u4` from `int_nvvm_mma_sp_m16n8k128_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.s8.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.u8.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k32_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k32.row.col.satfinite.u8` from `int_nvvm_mma_sp_m16n8k32_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e4m3_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e4m3.e4m3.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e4m3_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e4m3.e5m2.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e5m2_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e5m2.e4m3.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_f32_e5m2_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.f32.e5m2.e5m2.f32` from `int_nvvm_mma_sp_m16n8k64_row_col_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.4 on sm_89+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s4.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.s8.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u4.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u4` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.u8.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_m16n8k64_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.m16n8k64.row.col.satfinite.u8` from `int_nvvm_mma_sp_m16n8k64_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 7.1 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k128.row.col.satfinite.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k128_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k128.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f16_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.f16.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:2>, <register-list:2>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f32_bf16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.bf16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f32_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.f32.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k16_f32_tf32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k16.row.col.tf32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k16_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k16.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_f16_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.f16.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_f16_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f16.f16.f16.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_f32_bf16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.bf16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_bf16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f32.bf16.bf16.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_f32_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.f32.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_f32_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.f32.f16.f16.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k32.row.col.satfinite.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k32_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k32.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m1.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m1_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m1.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e2m3.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e2m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e2m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e3m2.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e3m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e3m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e4m3.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e4m3_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e4m3.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e2m1.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e2m1_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e2m1.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e2m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e2m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e2m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e3m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e3m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e3m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e4m3.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e4m3_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e4m3.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f16.e5m2.e5m2.f16` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f16_e5m2_e5m2_f16` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f16.e5m2.e5m2.f16 <register-list:2>, <register-list:4>, <register-list:4>, <register-list:2>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m1.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m1_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m1.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e2m3.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e2m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e2m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e3m2.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e3m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e3m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e4m3.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e4m3_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e4m3.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e2m1.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e2m1_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e2m1.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e2m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e2m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e2m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e3m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e3m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e3m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e4m3.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e4m3_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e4m3.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.kind.f8f6f4.f32.e5m2.e5m2.f32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_kind_f8f6f4_f32_e5m2_e5m2_f32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.7 on sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.kind::f8f6f4.f32.e5m2.e5m2.f32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s4.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s4_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.s8.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_s8_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.s8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4_s4` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4.s4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4_s4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.s4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u4` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u4` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u4.u4.s32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8_s8` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8.s8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8_s8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.s8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k64.row.col.satfinite.u8` from `int_nvvm_mma_sp_ordered_metadata_m16n8k64_row_col_satfinite_u8` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k64.row.col.satfinite.s32.u8.u8.s32 <register-list:4>, <register-list:4>, <register-list:4>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::mma_sp_ordered_metadata_m16n8k8_f32_tf32` | `nvvm.sparse_mma` | LLVM `llvm.nvvm.mma.sp.ordered.metadata.m16n8k8.row.col.tf32` from `int_nvvm_mma_sp_ordered_metadata_m16n8k8_row_col_tf32` | unsafe; scope warp; impure; memory none; convergent true | PTX 8.5 on sm_80+ ([PTX ISA 9.3, 9.7.15.6.3 Multiply-and-Accumulate Instruction: mma.sp](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma-sp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mma.sp::ordered_metadata.sync.aligned.m16n8k8.row.col.f32.tf32.tf32.f32 <register-list:4>, <register-list:2>, <register-list:2>, <register-list:4>, <register>, <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::movmatrix_trans_b16` | `nvvm.movmatrix_trans_b16` | PTX-native `movmatrix.sync.aligned.m8n8.trans.b16` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.8 on sm_75+ ([PTX ISA 9.3, 9.7.15.5.17 Warp-level matrix transpose instruction: movmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-movmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `movmatrix.sync.aligned.m8n8.trans.b16 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::mul_bf16x2` | `nvvm.mul_bf16x2` | PTX-native `mul.rn.bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.4.3 Half Precision Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::mul_f16x2` | `nvvm.mul_f16x2` | PTX-native `mul.rn.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.3 Half Precision Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::mul_f32x2` | `nvvm.mul_f32x2` | PTX-native `mul.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::mul_ftz_f32x2` | `nvvm.mul_ftz_f32x2` | PTX-native `mul.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.3 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.ftz.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rm_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rm.f` from `int_nvvm_mul_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rm.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rm_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rm.d` from `int_nvvm_mul_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rm.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rm_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rm.ftz.f` from `int_nvvm_mul_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rm.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rn_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rn.f` from `int_nvvm_mul_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rn_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rn.d` from `int_nvvm_mul_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rn_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rn.ftz.f` from `int_nvvm_mul_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rn.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rp_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rp.f` from `int_nvvm_mul_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rp.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rp_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rp.d` from `int_nvvm_mul_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rp.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rp_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rp.ftz.f` from `int_nvvm_mul_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rp.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rz.f` from `int_nvvm_mul_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rz_f64` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rz.d` from `int_nvvm_mul_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rz.f64 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::mul_rz_ftz_f32` | `nvvm.scalar_arithmetic` | LLVM `llvm.nvvm.mul.rz.ftz.f` from `int_nvvm_mul_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.5 Floating Point Instructions: mul](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mul.rz.ftz.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::barrier::nanosleep` | `nvvm.nanosleep` | LLVM `llvm.nvvm.nanosleep` from `int_nvvm_nanosleep` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 6.3 on sm_70+ ([PTX ISA 9.3, Parallel Synchronization and Communication Instructions: nanosleep](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-nanosleep)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `nanosleep.u32 <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::neg_bf16x2` | `nvvm.neg_bf16x2` | LLVM `llvm.nvvm.neg.bf16x2` from `int_nvvm_neg_bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.4.5 Half Precision Floating Point Instructions: neg](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `neg.bf16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::neg_f16x2` | `nvvm.neg_f16x2` | PTX-native `neg.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 6.0 on sm_70+ ([PTX ISA 9.3, 9.7.4.5 Half Precision Floating Point Instructions: neg](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `neg.f16x2 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::thread::nsmid` | `nvvm.read_ptx_sreg_nsmid` | LLVM `llvm.nvvm.read.ptx.sreg.nsmid` from `int_nvvm_read_ptx_sreg_nsmid` | safe; scope device; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.9 Special Registers: %nsmid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nsmid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nsmid;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::nwarpid` | `nvvm.read_ptx_sreg_nwarpid` | LLVM `llvm.nvvm.read.ptx.sreg.nwarpid` from `int_nvvm_read_ptx_sreg_nwarpid` | safe; scope cta; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 10.5 Special Registers: %nwarpid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-nwarpid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %nwarpid;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::atomic::atom_add_bf16x2` | `nvvm.packed_atomic_add` | PTX-native `atom.global.add.noftz.bf16x2` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.13.1.1 Atomic reduction and atomic operations: atom](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `atom.global.add.noftz.bf16x2 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::atomic::atom_add_f16x2` | `nvvm.packed_atomic_add` | PTX-native `atom.global.add.noftz.f16x2` | unsafe; scope thread; impure; memory read_write; convergent false | PTX 6.2 on sm_70+ ([PTX ISA 9.3, 9.7.13.1.1 Atomic reduction and atomic operations: atom](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `atom.global.add.noftz.f16x2 <register>, [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::debug::pmevent` | `nvvm.pmevent` | PTX-native `pmevent` | safe; scope thread; impure; memory none; convergent false | PTX 1.4 on all ([PTX ISA 9.3, 9.7.20.3 Miscellaneous Instructions: pmevent](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-pmevent)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `pmevent <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::prefetch_tma_descriptor` | `nvvm.prefetch_tensormap` | LLVM `llvm.nvvm.prefetch.tensormap` from `int_nvvm_prefetch_tensormap` | unsafe; scope thread; impure; memory read; convergent false | PTX 8.0 on sm_90+ ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `prefetch.tensormap [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt` from `int_nvvm_prmt` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32 <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt_b4e` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.b4e` from `int_nvvm_prmt_b4e` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32.b4e <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt_ecl` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.ecl` from `int_nvvm_prmt_ecl` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32.ecl <register>, <register>, 0, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt_ecr` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.ecr` from `int_nvvm_prmt_ecr` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32.ecr <register>, <register>, 0, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt_f4e` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.f4e` from `int_nvvm_prmt_f4e` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32.f4e <register>, <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt_rc16` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.rc16` from `int_nvvm_prmt_rc16` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32.rc16 <register>, <register>, 0, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::prmt::prmt_rc8` | `nvvm.prmt` | LLVM `llvm.nvvm.prmt.rc8` from `int_nvvm_prmt_rc8` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.9.7 Data Movement and Conversion Instructions: prmt](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-prmt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `prmt.b32.rc8 <register>, <register>, 0, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.approx.ftz.f` from `int_nvvm_rcp_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_approx_ftz_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.approx.ftz.d` from `int_nvvm_rcp_approx_ftz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.approx.ftz.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rm_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rm.f` from `int_nvvm_rcp_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rm.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rm_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rm.d` from `int_nvvm_rcp_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rm.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rm_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rm.ftz.f` from `int_nvvm_rcp_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rm.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rn_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rn.f` from `int_nvvm_rcp_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rn.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rn_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rn.d` from `int_nvvm_rcp_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rn.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rn_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rn.ftz.f` from `int_nvvm_rcp_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rn.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rp_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rp.f` from `int_nvvm_rcp_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rp.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rp_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rp.d` from `int_nvvm_rcp_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rp.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rp_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rp.ftz.f` from `int_nvvm_rcp_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rp.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rz.f` from `int_nvvm_rcp_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rz_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rz.d` from `int_nvvm_rcp_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rz.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rcp_rz_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rcp.rz.ftz.f` from `int_nvvm_rcp_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.7 Floating Point Instructions: rcp](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rcp)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rcp.rz.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_add` | `nvvm.redux_sync_add` | LLVM `llvm.nvvm.redux.sync.add` from `int_nvvm_redux_sync_add` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.add.s32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_and` | `nvvm.redux_sync_and` | LLVM `llvm.nvvm.redux.sync.and` from `int_nvvm_redux_sync_and` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.and.b32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_max_abs_f32` | `nvvm.redux_sync_fmax_abs` | LLVM `llvm.nvvm.redux.sync.fmax.abs` from `int_nvvm_redux_sync_fmax_abs` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.max.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_max_abs_nan_f32` | `nvvm.redux_sync_fmax_abs_nan` | LLVM `llvm.nvvm.redux.sync.fmax.abs.NaN` from `int_nvvm_redux_sync_fmax_abs_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.max.abs.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_max_f32` | `nvvm.redux_sync_fmax` | LLVM `llvm.nvvm.redux.sync.fmax` from `int_nvvm_redux_sync_fmax` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.max.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_max_i32` | `nvvm.redux_sync_max` | LLVM `llvm.nvvm.redux.sync.max` from `int_nvvm_redux_sync_max` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.max.s32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_max_nan_f32` | `nvvm.redux_sync_fmax_nan` | LLVM `llvm.nvvm.redux.sync.fmax.NaN` from `int_nvvm_redux_sync_fmax_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.max.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_max_u32` | `nvvm.redux_sync_umax` | LLVM `llvm.nvvm.redux.sync.umax` from `int_nvvm_redux_sync_umax` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.max.u32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_min_abs_f32` | `nvvm.redux_sync_fmin_abs` | LLVM `llvm.nvvm.redux.sync.fmin.abs` from `int_nvvm_redux_sync_fmin_abs` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.min.abs.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_min_abs_nan_f32` | `nvvm.redux_sync_fmin_abs_nan` | LLVM `llvm.nvvm.redux.sync.fmin.abs.NaN` from `int_nvvm_redux_sync_fmin_abs_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.min.abs.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_min_f32` | `nvvm.redux_sync_fmin` | LLVM `llvm.nvvm.redux.sync.fmin` from `int_nvvm_redux_sync_fmin` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.min.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_min_i32` | `nvvm.redux_sync_min` | LLVM `llvm.nvvm.redux.sync.min` from `int_nvvm_redux_sync_min` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.min.s32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_min_nan_f32` | `nvvm.redux_sync_fmin_nan` | LLVM `llvm.nvvm.redux.sync.fmin.NaN` from `int_nvvm_redux_sync_fmin_NaN` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.min.NaN.f32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_min_u32` | `nvvm.redux_sync_umin` | LLVM `llvm.nvvm.redux.sync.umin` from `int_nvvm_redux_sync_umin` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.min.u32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_or` | `nvvm.redux_sync_or` | LLVM `llvm.nvvm.redux.sync.or` from `int_nvvm_redux_sync_or` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.or.b32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::redux_sync_xor` | `nvvm.redux_sync_xor` | LLVM `llvm.nvvm.redux.sync.xor` from `int_nvvm_redux_sync_xor` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.14.13 Parallel Synchronization and Communication Instructions: redux.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-redux-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `redux.sync.xor.b32 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rsqrt_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rsqrt.approx.f` from `int_nvvm_rsqrt_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.14 Floating Point Instructions: rsqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rsqrt.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rsqrt_approx_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rsqrt.approx.d` from `int_nvvm_rsqrt_approx_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.14 Floating Point Instructions: rsqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rsqrt.approx.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::rsqrt_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.rsqrt.approx.ftz.f` from `int_nvvm_rsqrt_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.14 Floating Point Instructions: rsqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-rsqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `rsqrt.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::thread::setmaxnreg_dec` | `nvvm.setmaxnreg_dec` | LLVM `llvm.nvvm.setmaxnreg.dec.sync.aligned.u32` from `int_nvvm_setmaxnreg_dec_sync_aligned_u32` | unsafe; scope warpgroup; impure; memory none; convergent true | PTX 8.0 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, 9.7.16.5 Miscellaneous Instructions: setmaxnreg](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `setmaxnreg.dec.sync.aligned.u32 <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::thread::setmaxnreg_inc` | `nvvm.setmaxnreg_inc` | LLVM `llvm.nvvm.setmaxnreg.inc.sync.aligned.u32` from `int_nvvm_setmaxnreg_inc_sync_aligned_u32` | unsafe; scope warpgroup; impure; memory none; convergent true | PTX 8.0 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, 9.7.16.5 Miscellaneous Instructions: setmaxnreg](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg)) | validated on `rust-llvm-23.1.3-aab11ee0-execution-control`; expects `setmaxnreg.inc.sync.aligned.u32 <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_down_f32_sync` | `nvvm.shfl_sync_down_f32` | LLVM `llvm.nvvm.shfl.sync.down.f32` from `int_nvvm_shfl_sync_down_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.down.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_down_sync` | `nvvm.shfl_sync_down_i32` | LLVM `llvm.nvvm.shfl.sync.down.i32` from `int_nvvm_shfl_sync_down_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.down.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_down_u64_sync` | `nvvm.shfl_sync_down_i64` | PTX-native `shfl.sync.down.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.down.b32 lo, lo, <register>, 31, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_f32_sync` | `nvvm.shfl_sync_idx_f32` | LLVM `llvm.nvvm.shfl.sync.idx.f32` from `int_nvvm_shfl_sync_idx_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.idx.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_sync` | `nvvm.shfl_sync_idx_i32` | LLVM `llvm.nvvm.shfl.sync.idx.i32` from `int_nvvm_shfl_sync_idx_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.idx.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_u64_sync` | `nvvm.shfl_sync_idx_i64` | PTX-native `shfl.sync.idx.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.idx.b32 lo, lo, <register>, 31, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_up_f32_sync` | `nvvm.shfl_sync_up_f32` | LLVM `llvm.nvvm.shfl.sync.up.f32` from `int_nvvm_shfl_sync_up_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.up.b32 <register>, <register>, <register-or-immediate>, 0, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_up_sync` | `nvvm.shfl_sync_up_i32` | LLVM `llvm.nvvm.shfl.sync.up.i32` from `int_nvvm_shfl_sync_up_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.up.b32 <register>, <register>, <register-or-immediate>, 0, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_up_u64_sync` | `nvvm.shfl_sync_up_i64` | PTX-native `shfl.sync.up.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.up.b32 lo, lo, <register>, 0, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_xor_f32_sync` | `nvvm.shfl_sync_bfly_f32` | LLVM `llvm.nvvm.shfl.sync.bfly.f32` from `int_nvvm_shfl_sync_bfly_f32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.bfly.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_xor_sync` | `nvvm.shfl_sync_bfly_i32` | LLVM `llvm.nvvm.shfl.sync.bfly.i32` from `int_nvvm_shfl_sync_bfly_i32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.bfly.b32 <register>, <register>, <register-or-immediate>, 31, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::shuffle_xor_u64_sync` | `nvvm.shfl_sync_bfly_i64` | PTX-native `shfl.sync.bfly.b32` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.9.6 Data Movement and Conversion Instructions: shfl.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `shfl.sync.bfly.b32 lo, lo, <register>, 31, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sin_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sin.approx.f` from `int_nvvm_sin_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: sin](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sin)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sin.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sin_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sin.approx.ftz.f` from `int_nvvm_sin_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.10 Floating Point Instructions: sin](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sin)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sin.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::thread::smid` | `nvvm.read_ptx_sreg_smid` | LLVM `llvm.nvvm.read.ptx.sreg.smid` from `int_nvvm_read_ptx_sreg_smid` | safe; scope thread; impure; memory none; convergent false | PTX 1.3 on all ([PTX ISA 9.3, 10.8 Special Registers: %smid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-smid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %smid;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_approx_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.approx.f` from `int_nvvm_sqrt_approx_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_approx_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.approx.ftz.f` from `int_nvvm_sqrt_approx_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.approx.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rm_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rm.f` from `int_nvvm_sqrt_rm_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rm.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rm_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rm.d` from `int_nvvm_sqrt_rm_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rm.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rm_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rm.ftz.f` from `int_nvvm_sqrt_rm_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rm.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rn_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rn.f` from `int_nvvm_sqrt_rn_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rn.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rn_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rn.d` from `int_nvvm_sqrt_rn_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rn.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rn_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rn.ftz.f` from `int_nvvm_sqrt_rn_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rn.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rp_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rp.f` from `int_nvvm_sqrt_rp_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rp.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rp_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rp.d` from `int_nvvm_sqrt_rp_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rp.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rp_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rp.ftz.f` from `int_nvvm_sqrt_rp_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rp.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rz.f` from `int_nvvm_sqrt_rz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rz_f64` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rz.d` from `int_nvvm_sqrt_rz_d` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rz.f64 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::sqrt_rz_ftz_f32` | `nvvm.scalar_math` | LLVM `llvm.nvvm.sqrt.rz.ftz.f` from `int_nvvm_sqrt_rz_ftz_f` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.9 Floating Point Instructions: sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sqrt.rz.ftz.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::stmatrix_m8n8_x2_b16` | `nvvm.stmatrix_m8n8_x2` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x2_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `stmatrix.sync.aligned.m8n8.x2.shared.b16 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::stmatrix_m8n8_x2_trans_b16` | `nvvm.stmatrix_m8n8_x2_trans` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x2.trans.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x2_trans_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `stmatrix.sync.aligned.m8n8.x2.trans.shared.b16 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::stmatrix_m8n8_x4_b16` | `nvvm.stmatrix_m8n8_x4` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x4_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `stmatrix.sync.aligned.m8n8.x4.shared.b16 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::matrix::stmatrix_m8n8_x4_trans_b16` | `nvvm.stmatrix_m8n8_x4_trans` | LLVM `llvm.nvvm.stmatrix.sync.aligned.m8n8.x4.trans.b16` from `int_nvvm_stmatrix_sync_aligned_m8n8_x4_trans_b16` | unsafe; scope warp; impure; memory write; convergent true | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.14.5.16 Warp-level matrix store instruction: stmatrix](https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-stmatrix)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `stmatrix.sync.aligned.m8n8.x4.trans.shared.b16 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::bf16x2::sub_bf16x2` | `nvvm.sub_bf16x2` | PTX-native `sub.rn.bf16x2` | safe; scope thread; pure; memory none; convergent false | PTX 7.8 on sm_90+ ([PTX ISA 9.3, 9.7.4.2 Half Precision Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sub.rn.bf16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f16x2::sub_f16x2` | `nvvm.sub_f16x2` | PTX-native `sub.rn.f16x2` | safe; scope thread; pure; memory none; convergent false | PTX 4.2 on sm_70+ ([PTX ISA 9.3, 9.7.4.2 Half Precision Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sub.rn.f16x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::sub_f32x2` | `nvvm.sub_f32x2` | PTX-native `sub.rn.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.2 Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sub.rn.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::f32x2::sub_ftz_f32x2` | `nvvm.sub_ftz_f32x2` | PTX-native `sub.rn.ftz.f32x2` | safe; scope thread; pure; memory none; convergent false | PTX 8.6 on sm_100+ ([PTX ISA 9.3, 9.7.3.2 Floating Point Instructions: sub](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `sub.rn.ftz.f32x2 <register>, <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::sync_mask` | `nvvm.bar_warp_sync` | LLVM `llvm.nvvm.bar.warp.sync` from `int_nvvm_bar_warp_sync` | unsafe; scope warp; impure; memory read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.2 Parallel Synchronization and Communication Instructions: bar.warp.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-warp-sync)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `bar.warp.sync <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::thread::sync_threads` | `nvvm.barrier0` | LLVM `llvm.nvvm.barrier.cta.sync.aligned.all` from `int_nvvm_barrier_cta_sync_aligned_all` | unsafe; scope cta; impure; memory read_write; convergent true | PTX 1.0 on all ([PTX ISA 9.3, 9.7.14.1 Parallel Synchronization and Communication Instructions: bar, barrier](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `bar.sync 0;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::float::tanh_approx_f32` | `nvvm.scalar_math` | PTX-native `tanh.approx.f32` | safe; scope thread; pure; memory none; convergent false | PTX 7.0 on sm_80+ ([PTX ISA 9.3, 9.7.3.15 Floating Point Instructions: tanh](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-tanh)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tanh.approx.f32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_alloc` | `nvvm.tcgen05_alloc` | LLVM `llvm.nvvm.tcgen05.alloc.shared.cg1` from `int_nvvm_tcgen05_alloc_shared_cg1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.alloc.cta_group::1.sync.aligned.shared::cta.b32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_alloc_cg2` | `nvvm.tcgen05_alloc_cg2` | LLVM `llvm.nvvm.tcgen05.alloc.shared.cg2` from `int_nvvm_tcgen05_alloc_shared_cg2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.alloc.cta_group::2.sync.aligned.shared::cta.b32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_commit` | `nvvm.tcgen05_commit` | LLVM `llvm.nvvm.tcgen05.commit.cg1` from `int_nvvm_tcgen05_commit_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.commit.cta_group::1.mbarrier::arrive::one.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_commit_cg2` | `nvvm.tcgen05_commit_cg2` | LLVM `llvm.nvvm.tcgen05.commit.cg2` from `int_nvvm_tcgen05_commit_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.commit.cta_group::2.mbarrier::arrive::one.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_commit_multicast` | `nvvm.tcgen05_commit_multicast` | LLVM `llvm.nvvm.tcgen05.commit.mc.shared.cg1` from `int_nvvm_tcgen05_commit_mc_shared_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_commit_multicast_cg2` | `nvvm.tcgen05_commit_multicast_cg2` | LLVM `llvm.nvvm.tcgen05.commit.mc.shared.cg2` from `int_nvvm_tcgen05_commit_mc_shared_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.multicast::cluster.b64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_commit_shared_cluster` | `nvvm.tcgen05_commit_shared_cluster` | LLVM `llvm.nvvm.tcgen05.commit.shared.cg1` from `int_nvvm_tcgen05_commit_shared_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.commit.cta_group::1.mbarrier::arrive::one.shared::cluster.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_commit_shared_cluster_cg2` | `nvvm.tcgen05_commit_shared_cluster_cg2` | LLVM `llvm.nvvm.tcgen05.commit.shared.cg2` from `int_nvvm_tcgen05_commit_shared_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.commit.cta_group::2.mbarrier::arrive::one.shared::cluster.b64 [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b` | `nvvm.tcgen05_cp_128x128b` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.cg1` from `int_nvvm_tcgen05_cp_128x128b_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.128x128b [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b4x16_p64` | `nvvm.tcgen05_cp_128x128b_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_128x128b_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.128x128b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b4x16_p64_cg2` | `nvvm.tcgen05_cp_128x128b_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_128x128b_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.128x128b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b6x16_p32` | `nvvm.tcgen05_cp_128x128b_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_128x128b_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.128x128b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_b6x16_p32_cg2` | `nvvm.tcgen05_cp_128x128b_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_128x128b_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.128x128b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x128b_cg2` | `nvvm.tcgen05_cp_128x128b_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x128b.cg2` from `int_nvvm_tcgen05_cp_128x128b_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.128x128b [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b4x16_p64` | `nvvm.tcgen05_cp_128x256b_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_128x256b_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.128x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b4x16_p64_cg2` | `nvvm.tcgen05_cp_128x256b_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_128x256b_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.128x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b6x16_p32` | `nvvm.tcgen05_cp_128x256b_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_128x256b_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.128x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_128x256b_b6x16_p32_cg2` | `nvvm.tcgen05_cp_128x256b_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_128x256b_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.128x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4` | `nvvm.tcgen05_cp_32x128b_warpx4` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.cg1` from `int_nvvm_tcgen05_cp_32x128b_warpx4_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.32x128b.warpx4 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b4x16_p64` | `nvvm.tcgen05_cp_32x128b_warpx4_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.32x128b.warpx4.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2` | `nvvm.tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.32x128b.warpx4.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b6x16_p32` | `nvvm.tcgen05_cp_32x128b_warpx4_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.32x128b.warpx4.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2` | `nvvm.tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.32x128b.warpx4.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_32x128b_warpx4_cg2` | `nvvm.tcgen05_cp_32x128b_warpx4_cg2` | LLVM `llvm.nvvm.tcgen05.cp.32x128b_warpx4.cg2` from `int_nvvm_tcgen05_cp_32x128b_warpx4_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.32x128b.warpx4 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b` | `nvvm.tcgen05_cp_4x256b` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.cg1` from `int_nvvm_tcgen05_cp_4x256b_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.4x256b [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b4x16_p64` | `nvvm.tcgen05_cp_4x256b_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_4x256b_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.4x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b4x16_p64_cg2` | `nvvm.tcgen05_cp_4x256b_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_4x256b_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.4x256b.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b6x16_p32` | `nvvm.tcgen05_cp_4x256b_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_4x256b_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.4x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_b6x16_p32_cg2` | `nvvm.tcgen05_cp_4x256b_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_4x256b_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.4x256b.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_4x256b_cg2` | `nvvm.tcgen05_cp_4x256b_cg2` | LLVM `llvm.nvvm.tcgen05.cp.4x256b.cg2` from `int_nvvm_tcgen05_cp_4x256b_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.4x256b [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::01_23 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::01_23.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::01_23.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::01_23.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::01_23.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_01_23_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_01_23_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_01_23.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_01_23_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::01_23 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::02_13 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b4x16_p64.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::02_13.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b4x16_p64.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::02_13.b8x16.b4x16_p64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b6x16_p32.cg1` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.64x128b.warpx2::02_13.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.b6x16_p32.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::02_13.b8x16.b6x16_p32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_64x128b_warpx2_02_13_cg2` | `nvvm.tcgen05_cp_64x128b_warpx2_02_13_cg2` | LLVM `llvm.nvvm.tcgen05.cp.64x128b_warpx2_02_13.cg2` from `int_nvvm_tcgen05_cp_64x128b_warpx2_02_13_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.64x128b.warpx2::02_13 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_smem_to_tmem` | `nvvm.tcgen05_cp_smem_to_tmem` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.cg1` from `int_nvvm_tcgen05_cp_128x256b_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::1.128x256b [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_cp_smem_to_tmem_cg2` | `nvvm.tcgen05_cp_smem_to_tmem_cg2` | LLVM `llvm.nvvm.tcgen05.cp.128x256b.cg2` from `int_nvvm_tcgen05_cp_128x256b_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.cp.cta_group::2.128x256b [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_dealloc` | `nvvm.tcgen05_dealloc` | LLVM `llvm.nvvm.tcgen05.dealloc.cg1` from `int_nvvm_tcgen05_dealloc_cg1` | unsafe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.dealloc.cta_group::1.sync.aligned.b32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_dealloc_cg2` | `nvvm.tcgen05_dealloc_cg2` | LLVM `llvm.nvvm.tcgen05.dealloc.cg2` from `int_nvvm_tcgen05_dealloc_cg2` | unsafe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.dealloc.cta_group::2.sync.aligned.b32 <register>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_fence_after_thread_sync` | `nvvm.tcgen05_fence_after_thread_sync` | LLVM `llvm.nvvm.tcgen05.fence.after.thread.sync` from `int_nvvm_tcgen05_fence_after_thread_sync` | safe; scope thread; impure; memory none; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.fence::after_thread_sync;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_fence_before_thread_sync` | `nvvm.tcgen05_fence_before_thread_sync` | LLVM `llvm.nvvm.tcgen05.fence.before.thread.sync` from `int_nvvm_tcgen05_fence_before_thread_sync` | safe; scope thread; impure; memory none; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.fence::before_thread_sync;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x16_pack16` | `nvvm.tcgen05_ld_16x128b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x16` from `int_nvvm_tcgen05_ld_16x128b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x16.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x16_raw` | `nvvm.tcgen05_ld_16x128b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x16` from `int_nvvm_tcgen05_ld_16x128b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x16.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x1_pack16` | `nvvm.tcgen05_ld_16x128b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x1` from `int_nvvm_tcgen05_ld_16x128b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x1.pack::16b.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x1_raw` | `nvvm.tcgen05_ld_16x128b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x1` from `int_nvvm_tcgen05_ld_16x128b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x1.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x2_pack16` | `nvvm.tcgen05_ld_16x128b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x2` from `int_nvvm_tcgen05_ld_16x128b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x2.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x2_raw` | `nvvm.tcgen05_ld_16x128b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x2` from `int_nvvm_tcgen05_ld_16x128b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x2.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x32_pack16` | `nvvm.tcgen05_ld_16x128b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x32` from `int_nvvm_tcgen05_ld_16x128b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x32.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x32_raw` | `nvvm.tcgen05_ld_16x128b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x32` from `int_nvvm_tcgen05_ld_16x128b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x32.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x4_pack16` | `nvvm.tcgen05_ld_16x128b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x4` from `int_nvvm_tcgen05_ld_16x128b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x4.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x4_raw` | `nvvm.tcgen05_ld_16x128b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x4` from `int_nvvm_tcgen05_ld_16x128b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x4.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x64_pack16` | `nvvm.tcgen05_ld_16x128b_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x64` from `int_nvvm_tcgen05_ld_16x128b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x64.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x64_raw` | `nvvm.tcgen05_ld_16x128b_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x64` from `int_nvvm_tcgen05_ld_16x128b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x64.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x8_pack16` | `nvvm.tcgen05_ld_16x128b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x8` from `int_nvvm_tcgen05_ld_16x128b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x8.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x128b_x8_raw` | `nvvm.tcgen05_ld_16x128b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x128b.x8` from `int_nvvm_tcgen05_ld_16x128b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x128b.x8.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_pure` | `nvvm.tcgen05_ld_16x256b_pure` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x1` from `int_nvvm_tcgen05_ld_16x256b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x1.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x16_pack16` | `nvvm.tcgen05_ld_16x256b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x16` from `int_nvvm_tcgen05_ld_16x256b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x16.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x16_raw` | `nvvm.tcgen05_ld_16x256b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x16` from `int_nvvm_tcgen05_ld_16x256b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x16.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x1_pack16` | `nvvm.tcgen05_ld_16x256b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x1` from `int_nvvm_tcgen05_ld_16x256b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x1.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x1_raw` | `nvvm.tcgen05_ld_16x256b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x1` from `int_nvvm_tcgen05_ld_16x256b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x1.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x2_pack16` | `nvvm.tcgen05_ld_16x256b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x2` from `int_nvvm_tcgen05_ld_16x256b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x2.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x2_raw` | `nvvm.tcgen05_ld_16x256b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x2` from `int_nvvm_tcgen05_ld_16x256b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x2.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x32_pack16` | `nvvm.tcgen05_ld_16x256b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x32` from `int_nvvm_tcgen05_ld_16x256b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x32.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x32_raw` | `nvvm.tcgen05_ld_16x256b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x32` from `int_nvvm_tcgen05_ld_16x256b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x32.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x4_pack16` | `nvvm.tcgen05_ld_16x256b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x4` from `int_nvvm_tcgen05_ld_16x256b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x4.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x4_raw` | `nvvm.tcgen05_ld_16x256b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x4` from `int_nvvm_tcgen05_ld_16x256b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x4.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x8_pack16` | `nvvm.tcgen05_ld_16x256b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x8` from `int_nvvm_tcgen05_ld_16x256b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x8.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x8_pure` | `nvvm.tcgen05_ld_16x256b_x8_pure` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x8` from `int_nvvm_tcgen05_ld_16x256b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x8.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x256b_x8_raw` | `nvvm.tcgen05_ld_16x256b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x256b.x8` from `int_nvvm_tcgen05_ld_16x256b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x256b.x8.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x128_pack16` | `nvvm.tcgen05_ld_16x32bx2_x128_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x128` from `int_nvvm_tcgen05_ld_16x32bx2_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x128.pack::16b.b32 <register-list:128>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x128_raw` | `nvvm.tcgen05_ld_16x32bx2_x128_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x128` from `int_nvvm_tcgen05_ld_16x32bx2_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x128.b32 <register-list:128>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x16_pack16` | `nvvm.tcgen05_ld_16x32bx2_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x16` from `int_nvvm_tcgen05_ld_16x32bx2_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x16.pack::16b.b32 <register-list:16>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x16_raw` | `nvvm.tcgen05_ld_16x32bx2_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x16` from `int_nvvm_tcgen05_ld_16x32bx2_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x16.b32 <register-list:16>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x1_pack16` | `nvvm.tcgen05_ld_16x32bx2_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x1` from `int_nvvm_tcgen05_ld_16x32bx2_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x1.pack::16b.b32 <register-list:1>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x1_raw` | `nvvm.tcgen05_ld_16x32bx2_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x1` from `int_nvvm_tcgen05_ld_16x32bx2_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x1.b32 <register-list:1>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x2_pack16` | `nvvm.tcgen05_ld_16x32bx2_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x2` from `int_nvvm_tcgen05_ld_16x32bx2_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x2.pack::16b.b32 <register-list:2>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x2_raw` | `nvvm.tcgen05_ld_16x32bx2_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x2` from `int_nvvm_tcgen05_ld_16x32bx2_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x2.b32 <register-list:2>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x32_pack16` | `nvvm.tcgen05_ld_16x32bx2_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x32` from `int_nvvm_tcgen05_ld_16x32bx2_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x32.pack::16b.b32 <register-list:32>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x32_raw` | `nvvm.tcgen05_ld_16x32bx2_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x32` from `int_nvvm_tcgen05_ld_16x32bx2_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x32.b32 <register-list:32>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x4_pack16` | `nvvm.tcgen05_ld_16x32bx2_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x4` from `int_nvvm_tcgen05_ld_16x32bx2_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x4.pack::16b.b32 <register-list:4>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x4_raw` | `nvvm.tcgen05_ld_16x32bx2_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x4` from `int_nvvm_tcgen05_ld_16x32bx2_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x4.b32 <register-list:4>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x64_pack16` | `nvvm.tcgen05_ld_16x32bx2_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x64` from `int_nvvm_tcgen05_ld_16x32bx2_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x64.pack::16b.b32 <register-list:64>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x64_raw` | `nvvm.tcgen05_ld_16x32bx2_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x64` from `int_nvvm_tcgen05_ld_16x32bx2_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x64.b32 <register-list:64>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x8_pack16` | `nvvm.tcgen05_ld_16x32bx2_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x8` from `int_nvvm_tcgen05_ld_16x32bx2_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x8.pack::16b.b32 <register-list:8>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x32bx2_x8_raw` | `nvvm.tcgen05_ld_16x32bx2_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x32bx2.x8` from `int_nvvm_tcgen05_ld_16x32bx2_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x32bx2.x8.b32 <register-list:8>, [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x128_pack16` | `nvvm.tcgen05_ld_16x64b_x128_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x128` from `int_nvvm_tcgen05_ld_16x64b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x128.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x128_raw` | `nvvm.tcgen05_ld_16x64b_x128_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x128` from `int_nvvm_tcgen05_ld_16x64b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x128.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x16_pack16` | `nvvm.tcgen05_ld_16x64b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x16` from `int_nvvm_tcgen05_ld_16x64b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x16.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x16_raw` | `nvvm.tcgen05_ld_16x64b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x16` from `int_nvvm_tcgen05_ld_16x64b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x16.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x1_pack16` | `nvvm.tcgen05_ld_16x64b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x1` from `int_nvvm_tcgen05_ld_16x64b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x1.pack::16b.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x1_raw` | `nvvm.tcgen05_ld_16x64b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x1` from `int_nvvm_tcgen05_ld_16x64b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x1.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x2_pack16` | `nvvm.tcgen05_ld_16x64b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x2` from `int_nvvm_tcgen05_ld_16x64b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x2.pack::16b.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x2_raw` | `nvvm.tcgen05_ld_16x64b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x2` from `int_nvvm_tcgen05_ld_16x64b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x2.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x32_pack16` | `nvvm.tcgen05_ld_16x64b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x32` from `int_nvvm_tcgen05_ld_16x64b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x32.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x32_raw` | `nvvm.tcgen05_ld_16x64b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x32` from `int_nvvm_tcgen05_ld_16x64b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x32.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x4_pack16` | `nvvm.tcgen05_ld_16x64b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x4` from `int_nvvm_tcgen05_ld_16x64b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x4.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x4_raw` | `nvvm.tcgen05_ld_16x64b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x4` from `int_nvvm_tcgen05_ld_16x64b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x4.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x64_pack16` | `nvvm.tcgen05_ld_16x64b_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x64` from `int_nvvm_tcgen05_ld_16x64b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x64.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x64_raw` | `nvvm.tcgen05_ld_16x64b_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x64` from `int_nvvm_tcgen05_ld_16x64b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x64.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x8_pack16` | `nvvm.tcgen05_ld_16x64b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x8` from `int_nvvm_tcgen05_ld_16x64b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x8.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_16x64b_x8_raw` | `nvvm.tcgen05_ld_16x64b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.16x64b.x8` from `int_nvvm_tcgen05_ld_16x64b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.16x64b.x8.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x128_pack16` | `nvvm.tcgen05_ld_32x32b_x128_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x128` from `int_nvvm_tcgen05_ld_32x32b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x128.pack::16b.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x128_raw` | `nvvm.tcgen05_ld_32x32b_x128_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x128` from `int_nvvm_tcgen05_ld_32x32b_x128` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x128.b32 <register-list:128>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x16_pack16` | `nvvm.tcgen05_ld_32x32b_x16_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x16` from `int_nvvm_tcgen05_ld_32x32b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x16.pack::16b.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x16_raw` | `nvvm.tcgen05_ld_32x32b_x16_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x16` from `int_nvvm_tcgen05_ld_32x32b_x16` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x16.b32 <register-list:16>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x1_pack16` | `nvvm.tcgen05_ld_32x32b_x1_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x1` from `int_nvvm_tcgen05_ld_32x32b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x1.pack::16b.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x1_raw` | `nvvm.tcgen05_ld_32x32b_x1_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x1` from `int_nvvm_tcgen05_ld_32x32b_x1` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x1.b32 <register-list:1>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x2_pack16` | `nvvm.tcgen05_ld_32x32b_x2_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x2` from `int_nvvm_tcgen05_ld_32x32b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x2.pack::16b.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x2_raw` | `nvvm.tcgen05_ld_32x32b_x2_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x2` from `int_nvvm_tcgen05_ld_32x32b_x2` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x2.b32 <register-list:2>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x32_pack16` | `nvvm.tcgen05_ld_32x32b_x32_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x32` from `int_nvvm_tcgen05_ld_32x32b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x32.pack::16b.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x32_raw` | `nvvm.tcgen05_ld_32x32b_x32_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x32` from `int_nvvm_tcgen05_ld_32x32b_x32` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x32.b32 <register-list:32>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x4_pack16` | `nvvm.tcgen05_ld_32x32b_x4_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x4` from `int_nvvm_tcgen05_ld_32x32b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x4.pack::16b.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x4_raw` | `nvvm.tcgen05_ld_32x32b_x4_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x4` from `int_nvvm_tcgen05_ld_32x32b_x4` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x4.b32 <register-list:4>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x64_pack16` | `nvvm.tcgen05_ld_32x32b_x64_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x64` from `int_nvvm_tcgen05_ld_32x32b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x64.pack::16b.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x64_raw` | `nvvm.tcgen05_ld_32x32b_x64_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x64` from `int_nvvm_tcgen05_ld_32x32b_x64` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x64.b32 <register-list:64>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x8_pack16` | `nvvm.tcgen05_ld_32x32b_x8_pack16` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x8` from `int_nvvm_tcgen05_ld_32x32b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x8.pack::16b.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_ld_32x32b_x8_raw` | `nvvm.tcgen05_ld_32x32b_x8_raw` | LLVM `llvm.nvvm.tcgen05.ld.32x32b.x8` from `int_nvvm_tcgen05_ld_32x32b_x8` | unsafe; scope warp; impure; memory read; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.ld.sync.aligned.32x32b.x8.b32 <register-list:8>, [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_load_wait` | `nvvm.tcgen05_load_wait` | LLVM `llvm.nvvm.tcgen05.wait.ld` from `int_nvvm_tcgen05_wait_ld` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.wait::ld.sync.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_e2m1` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_e2m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_e3m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_e4m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_e5m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_f16` | `nvvm.tcgen05_mma_f16` | LLVM `llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1` from `int_nvvm_tcgen05_mma_shared_disable_output_lane_cg1` | unsafe; scope thread; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.mma.cta_group::1.kind::f16 [<address>], <register>, <register>, <register>, {%z, %z, %z, %z}, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_f16_cg2` | `nvvm.tcgen05_mma_f16_cg2` | LLVM `llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2` from `int_nvvm_tcgen05_mma_shared_disable_output_lane_cg2` | unsafe; scope thread; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.mma.cta_group::2.kind::f16 [<address>], <register>, <register>, <register>, {%z, %z, %z, %z, %z, %z, %z, %z}, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.shared` from `int_nvvm_tcgen05_mma_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_sp_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.sp.shared` from `int_nvvm_tcgen05_mma_sp_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [<address>], <register>, <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_sp_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.sp.tensor` from `int_nvvm_tcgen05_mma_sp_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [<address>], [<address>], <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_sp_tensor_ashift` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.sp.tensor.ashift` from `int_nvvm_tcgen05_mma_sp_tensor_ashift` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.ashift [<address>], [<address>], <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.tensor` from `int_nvvm_tcgen05_mma_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_tensor_ashift` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.tensor.ashift` from `int_nvvm_tcgen05_mma_tensor_ashift` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.ashift [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_bf16` | `nvvm.tcgen05_mma_ws_bf16` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.mma.ws.cta_group::1.kind::f16 [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e2m1` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e2m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e3m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e4m3` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_e5m2` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f8f6f4.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_f16` | `nvvm.tcgen05_mma_ws_f16` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.mma.ws.cta_group::1.kind::f16 [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.shared` from `int_nvvm_tcgen05_mma_ws_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_shared_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.shared.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_shared_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, <register>, %enable_pred, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_shared` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.shared` from `int_nvvm_tcgen05_mma_ws_sp_shared` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_shared_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.shared.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_sp_shared_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], <register>, <register>, [<address>], <register>, %enable_pred, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.tensor` from `int_nvvm_tcgen05_mma_ws_sp_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, [<address>], <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_sp_tensor_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.sp.tensor.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_sp_tensor_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.sp.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, [<address>], <register>, %enable_pred, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_tensor` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_tensor_zero_col_mask` | `nvvm.tcgen05_mma` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor.zero_col_mask` from `int_nvvm_tcgen05_mma_ws_tensor_zero_col_mask` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32 ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | lowered on `rust-llvm-23.1.3-aab11ee0-tcgen05-mma`; expects `tcgen05.mma.ws.cta_group::1.kind::f16.collector::b0::discard [<address>], [<address>], <register>, <register>, %enable_pred, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_mma_ws_tf32` | `nvvm.tcgen05_mma_ws_tf32` | LLVM `llvm.nvvm.tcgen05.mma.ws.tensor` from `int_nvvm_tcgen05_mma_ws_tensor` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.mma.ws.cta_group::1.kind::tf32 [<address>], [<address>], <register>, <register>, %enable_pred;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_relinquish_alloc_permit` | `nvvm.tcgen05_relinquish_alloc_permit` | LLVM `llvm.nvvm.tcgen05.relinq.alloc.permit.cg1` from `int_nvvm_tcgen05_relinq_alloc_permit_cg1` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.relinquish_alloc_permit.cta_group::1.sync.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_relinquish_alloc_permit_cg2` | `nvvm.tcgen05_relinquish_alloc_permit_cg2` | LLVM `llvm.nvvm.tcgen05.relinq.alloc.permit.cg2` from `int_nvvm_tcgen05_relinq_alloc_permit_cg2` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.relinquish_alloc_permit.cta_group::2.sync.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_shift_down` | `nvvm.tcgen05_shift_down` | LLVM `llvm.nvvm.tcgen05.shift.down.cg1` from `int_nvvm_tcgen05_shift_down_cg1` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.shift.cta_group::1.down [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_shift_down_cg2` | `nvvm.tcgen05_shift_down_cg2` | LLVM `llvm.nvvm.tcgen05.shift.down.cg2` from `int_nvvm_tcgen05_shift_down_cg2` | unsafe; scope thread; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.shift.cta_group::2.down [<address>];` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x16_raw` | `nvvm.tcgen05_st_16x128b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x16` from `int_nvvm_tcgen05_st_16x128b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x16.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x16_unpack16` | `nvvm.tcgen05_st_16x128b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x16` from `int_nvvm_tcgen05_st_16x128b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x16.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x1_raw` | `nvvm.tcgen05_st_16x128b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x1` from `int_nvvm_tcgen05_st_16x128b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x1.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x1_unpack16` | `nvvm.tcgen05_st_16x128b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x1` from `int_nvvm_tcgen05_st_16x128b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x1.unpack::16b.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x2_raw` | `nvvm.tcgen05_st_16x128b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x2` from `int_nvvm_tcgen05_st_16x128b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x2.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x2_unpack16` | `nvvm.tcgen05_st_16x128b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x2` from `int_nvvm_tcgen05_st_16x128b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x2.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x32_raw` | `nvvm.tcgen05_st_16x128b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x32` from `int_nvvm_tcgen05_st_16x128b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x32.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x32_unpack16` | `nvvm.tcgen05_st_16x128b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x32` from `int_nvvm_tcgen05_st_16x128b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x32.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x4_raw` | `nvvm.tcgen05_st_16x128b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x4` from `int_nvvm_tcgen05_st_16x128b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x4.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x4_unpack16` | `nvvm.tcgen05_st_16x128b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x4` from `int_nvvm_tcgen05_st_16x128b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x4.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x64_raw` | `nvvm.tcgen05_st_16x128b_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x64` from `int_nvvm_tcgen05_st_16x128b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x64.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x64_unpack16` | `nvvm.tcgen05_st_16x128b_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x64` from `int_nvvm_tcgen05_st_16x128b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x64.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x8_raw` | `nvvm.tcgen05_st_16x128b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x8` from `int_nvvm_tcgen05_st_16x128b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x8.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x128b_x8_unpack16` | `nvvm.tcgen05_st_16x128b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x128b.x8` from `int_nvvm_tcgen05_st_16x128b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x128b.x8.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x16_raw` | `nvvm.tcgen05_st_16x256b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x16` from `int_nvvm_tcgen05_st_16x256b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x16.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x16_unpack16` | `nvvm.tcgen05_st_16x256b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x16` from `int_nvvm_tcgen05_st_16x256b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x16.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x1_raw` | `nvvm.tcgen05_st_16x256b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x1` from `int_nvvm_tcgen05_st_16x256b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x1.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x1_unpack16` | `nvvm.tcgen05_st_16x256b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x1` from `int_nvvm_tcgen05_st_16x256b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x1.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x2_raw` | `nvvm.tcgen05_st_16x256b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x2` from `int_nvvm_tcgen05_st_16x256b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x2.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x2_unpack16` | `nvvm.tcgen05_st_16x256b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x2` from `int_nvvm_tcgen05_st_16x256b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x2.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x32_raw` | `nvvm.tcgen05_st_16x256b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x32` from `int_nvvm_tcgen05_st_16x256b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x32.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x32_unpack16` | `nvvm.tcgen05_st_16x256b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x32` from `int_nvvm_tcgen05_st_16x256b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x32.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x4_raw` | `nvvm.tcgen05_st_16x256b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x4` from `int_nvvm_tcgen05_st_16x256b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x4.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x4_unpack16` | `nvvm.tcgen05_st_16x256b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x4` from `int_nvvm_tcgen05_st_16x256b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x4.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x8_raw` | `nvvm.tcgen05_st_16x256b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x8` from `int_nvvm_tcgen05_st_16x256b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x8.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x256b_x8_unpack16` | `nvvm.tcgen05_st_16x256b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x256b.x8` from `int_nvvm_tcgen05_st_16x256b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x256b.x8.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x128_raw` | `nvvm.tcgen05_st_16x32bx2_x128_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x128` from `int_nvvm_tcgen05_st_16x32bx2_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x128.b32 [<address>], <immediate>, <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x128_unpack16` | `nvvm.tcgen05_st_16x32bx2_x128_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x128` from `int_nvvm_tcgen05_st_16x32bx2_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x128.unpack::16b.b32 [<address>], <immediate>, <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x16_raw` | `nvvm.tcgen05_st_16x32bx2_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x16` from `int_nvvm_tcgen05_st_16x32bx2_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x16.b32 [<address>], <immediate>, <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x16_unpack16` | `nvvm.tcgen05_st_16x32bx2_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x16` from `int_nvvm_tcgen05_st_16x32bx2_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x16.unpack::16b.b32 [<address>], <immediate>, <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x1_raw` | `nvvm.tcgen05_st_16x32bx2_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x1` from `int_nvvm_tcgen05_st_16x32bx2_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x1.b32 [<address>], <immediate>, <register-list:1>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x1_unpack16` | `nvvm.tcgen05_st_16x32bx2_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x1` from `int_nvvm_tcgen05_st_16x32bx2_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x1.unpack::16b.b32 [<address>], <immediate>, <register-list:1>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x2_raw` | `nvvm.tcgen05_st_16x32bx2_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x2` from `int_nvvm_tcgen05_st_16x32bx2_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x2.b32 [<address>], <immediate>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x2_unpack16` | `nvvm.tcgen05_st_16x32bx2_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x2` from `int_nvvm_tcgen05_st_16x32bx2_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x2.unpack::16b.b32 [<address>], <immediate>, <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x32_raw` | `nvvm.tcgen05_st_16x32bx2_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x32` from `int_nvvm_tcgen05_st_16x32bx2_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x32.b32 [<address>], <immediate>, <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x32_unpack16` | `nvvm.tcgen05_st_16x32bx2_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x32` from `int_nvvm_tcgen05_st_16x32bx2_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x32.unpack::16b.b32 [<address>], <immediate>, <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x4_raw` | `nvvm.tcgen05_st_16x32bx2_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x4` from `int_nvvm_tcgen05_st_16x32bx2_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x4.b32 [<address>], <immediate>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x4_unpack16` | `nvvm.tcgen05_st_16x32bx2_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x4` from `int_nvvm_tcgen05_st_16x32bx2_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x4.unpack::16b.b32 [<address>], <immediate>, <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x64_raw` | `nvvm.tcgen05_st_16x32bx2_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x64` from `int_nvvm_tcgen05_st_16x32bx2_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x64.b32 [<address>], <immediate>, <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x64_unpack16` | `nvvm.tcgen05_st_16x32bx2_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x64` from `int_nvvm_tcgen05_st_16x32bx2_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x64.unpack::16b.b32 [<address>], <immediate>, <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x8_raw` | `nvvm.tcgen05_st_16x32bx2_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x8` from `int_nvvm_tcgen05_st_16x32bx2_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x8.b32 [<address>], <immediate>, <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x32bx2_x8_unpack16` | `nvvm.tcgen05_st_16x32bx2_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x32bx2.x8` from `int_nvvm_tcgen05_st_16x32bx2_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x32bx2.x8.unpack::16b.b32 [<address>], <immediate>, <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x128_raw` | `nvvm.tcgen05_st_16x64b_x128_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x128` from `int_nvvm_tcgen05_st_16x64b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x128.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x128_unpack16` | `nvvm.tcgen05_st_16x64b_x128_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x128` from `int_nvvm_tcgen05_st_16x64b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x128.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x16_raw` | `nvvm.tcgen05_st_16x64b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x16` from `int_nvvm_tcgen05_st_16x64b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x16.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x16_unpack16` | `nvvm.tcgen05_st_16x64b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x16` from `int_nvvm_tcgen05_st_16x64b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x16.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x1_raw` | `nvvm.tcgen05_st_16x64b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x1` from `int_nvvm_tcgen05_st_16x64b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x1.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x1_unpack16` | `nvvm.tcgen05_st_16x64b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x1` from `int_nvvm_tcgen05_st_16x64b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x1.unpack::16b.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x2_raw` | `nvvm.tcgen05_st_16x64b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x2` from `int_nvvm_tcgen05_st_16x64b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x2.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x2_unpack16` | `nvvm.tcgen05_st_16x64b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x2` from `int_nvvm_tcgen05_st_16x64b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x2.unpack::16b.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x32_raw` | `nvvm.tcgen05_st_16x64b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x32` from `int_nvvm_tcgen05_st_16x64b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x32.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x32_unpack16` | `nvvm.tcgen05_st_16x64b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x32` from `int_nvvm_tcgen05_st_16x64b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x32.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x4_raw` | `nvvm.tcgen05_st_16x64b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x4` from `int_nvvm_tcgen05_st_16x64b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x4.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x4_unpack16` | `nvvm.tcgen05_st_16x64b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x4` from `int_nvvm_tcgen05_st_16x64b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x4.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x64_raw` | `nvvm.tcgen05_st_16x64b_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x64` from `int_nvvm_tcgen05_st_16x64b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x64.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x64_unpack16` | `nvvm.tcgen05_st_16x64b_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x64` from `int_nvvm_tcgen05_st_16x64b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x64.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x8_raw` | `nvvm.tcgen05_st_16x64b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x8` from `int_nvvm_tcgen05_st_16x64b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x8.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_16x64b_x8_unpack16` | `nvvm.tcgen05_st_16x64b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.16x64b.x8` from `int_nvvm_tcgen05_st_16x64b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.16x64b.x8.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x128_raw` | `nvvm.tcgen05_st_32x32b_x128_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x128` from `int_nvvm_tcgen05_st_32x32b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x128.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x128_unpack16` | `nvvm.tcgen05_st_32x32b_x128_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x128` from `int_nvvm_tcgen05_st_32x32b_x128` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x128.unpack::16b.b32 [<address>], <register-list:128>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x16_raw` | `nvvm.tcgen05_st_32x32b_x16_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x16` from `int_nvvm_tcgen05_st_32x32b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x16.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x16_unpack16` | `nvvm.tcgen05_st_32x32b_x16_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x16` from `int_nvvm_tcgen05_st_32x32b_x16` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x16.unpack::16b.b32 [<address>], <register-list:16>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x1_raw` | `nvvm.tcgen05_st_32x32b_x1_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x1` from `int_nvvm_tcgen05_st_32x32b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x1.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x1_unpack16` | `nvvm.tcgen05_st_32x32b_x1_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x1` from `int_nvvm_tcgen05_st_32x32b_x1` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x1.unpack::16b.b32 [<address>], <register-list:1>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x2_raw` | `nvvm.tcgen05_st_32x32b_x2_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x2` from `int_nvvm_tcgen05_st_32x32b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x2.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x2_unpack16` | `nvvm.tcgen05_st_32x32b_x2_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x2` from `int_nvvm_tcgen05_st_32x32b_x2` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x2.unpack::16b.b32 [<address>], <register-list:2>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x32_raw` | `nvvm.tcgen05_st_32x32b_x32_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x32` from `int_nvvm_tcgen05_st_32x32b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x32.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x32_unpack16` | `nvvm.tcgen05_st_32x32b_x32_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x32` from `int_nvvm_tcgen05_st_32x32b_x32` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x32.unpack::16b.b32 [<address>], <register-list:32>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x4_raw` | `nvvm.tcgen05_st_32x32b_x4_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x4` from `int_nvvm_tcgen05_st_32x32b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x4.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x4_unpack16` | `nvvm.tcgen05_st_32x32b_x4_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x4` from `int_nvvm_tcgen05_st_32x32b_x4` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x4.unpack::16b.b32 [<address>], <register-list:4>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x64_raw` | `nvvm.tcgen05_st_32x32b_x64_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x64` from `int_nvvm_tcgen05_st_32x32b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x64.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x64_unpack16` | `nvvm.tcgen05_st_32x32b_x64_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x64` from `int_nvvm_tcgen05_st_32x32b_x64` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x64.unpack::16b.b32 [<address>], <register-list:64>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x8_raw` | `nvvm.tcgen05_st_32x32b_x8_raw` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x8` from `int_nvvm_tcgen05_st_32x32b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x8.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_st_32x32b_x8_unpack16` | `nvvm.tcgen05_st_32x32b_x8_unpack16` | LLVM `llvm.nvvm.tcgen05.st.32x32b.x8` from `int_nvvm_tcgen05_st_32x32b_x8` | unsafe; scope warp; impure; memory write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.st.sync.aligned.32x32b.x8.unpack::16b.b32 [<address>], <register-list:8>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tcgen05::tcgen05_store_wait` | `nvvm.tcgen05_store_wait` | LLVM `llvm.nvvm.tcgen05.wait.st` from `int_nvvm_tcgen05_wait_st` | safe; scope warp; impure; memory read_write; convergent true | PTX 8.6 on sm_100a or sm_101a or sm_103a or sm_110a ([PTX ISA 8.6, Tensor Memory tcgen05 instructions](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-memory)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `tcgen05.wait::st.sync.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_box_dim` | `nvvm.tensormap_replace_box_dim` | LLVM `llvm.nvvm.tensormap.replace.box.dim` from `int_nvvm_tensormap_replace_box_dim` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.box_dim.global.b1024.b32 [<address>], <immediate>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_element_stride` | `nvvm.tensormap_replace_element_stride` | LLVM `llvm.nvvm.tensormap.replace.element.stride` from `int_nvvm_tensormap_replace_element_stride` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.element_stride.global.b1024.b32 [<address>], <immediate>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_element_type` | `nvvm.tensormap_replace_element_type` | LLVM `llvm.nvvm.tensormap.replace.elemtype` from `int_nvvm_tensormap_replace_elemtype` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.elemtype.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_fill_mode` | `nvvm.tensormap_replace_fill_mode` | LLVM `llvm.nvvm.tensormap.replace.fill.mode` from `int_nvvm_tensormap_replace_fill_mode` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.fill_mode.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_global_address` | `nvvm.tensormap_replace_global_address` | LLVM `llvm.nvvm.tensormap.replace.global.address` from `int_nvvm_tensormap_replace_global_address` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.global_address.global.b1024.b64 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_global_dim` | `nvvm.tensormap_replace_global_dim` | LLVM `llvm.nvvm.tensormap.replace.global.dim` from `int_nvvm_tensormap_replace_global_dim` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.global_dim.global.b1024.b32 [<address>], <immediate>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_global_stride` | `nvvm.tensormap_replace_global_stride` | LLVM `llvm.nvvm.tensormap.replace.global.stride` from `int_nvvm_tensormap_replace_global_stride` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.global_stride.global.b1024.b64 [<address>], <immediate>, <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_interleave_layout` | `nvvm.tensormap_replace_interleave_layout` | LLVM `llvm.nvvm.tensormap.replace.interleave.layout` from `int_nvvm_tensormap_replace_interleave_layout` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.interleave_layout.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_rank` | `nvvm.tensormap_replace_rank` | LLVM `llvm.nvvm.tensormap.replace.rank` from `int_nvvm_tensormap_replace_rank` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.rank.global.b1024.b32 [<address>], <register>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_swizzle_atomicity` | `nvvm.tensormap_replace_swizzle_atomicity` | LLVM `llvm.nvvm.tensormap.replace.swizzle.atomicity` from `int_nvvm_tensormap_replace_swizzle_atomicity` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.6 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.swizzle_atomicity.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::tma::tensormap_replace_swizzle_mode` | `nvvm.tensormap_replace_swizzle_mode` | LLVM `llvm.nvvm.tensormap.replace.swizzle.mode` from `int_nvvm_tensormap_replace_swizzle_mode` | unsafe; scope thread; impure; memory write; convergent false | PTX 8.3 on sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a ([PTX ISA 9.3, Tensor-map descriptor and asynchronous bulk tensor operations](https://docs.nvidia.com/cuda/parallel-thread-execution/#tensor-map)) | validated on `rust-llvm-23.1.3-aab11ee0-tma`; expects `tensormap.replace.tile.swizzle_mode.global.b1024.b32 [<address>], <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::thread_idx_x` | `nvvm.read_ptx_sreg_tid_x` | LLVM `llvm.nvvm.read.ptx.sreg.tid.x` from `int_nvvm_read_ptx_sreg_tid_x` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.1 Special Registers: %tid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %tid.x;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::thread_idx_y` | `nvvm.read_ptx_sreg_tid_y` | LLVM `llvm.nvvm.read.ptx.sreg.tid.y` from `int_nvvm_read_ptx_sreg_tid_y` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.1 Special Registers: %tid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %tid.y;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::sreg::thread_idx_z` | `nvvm.read_ptx_sreg_tid_z` | LLVM `llvm.nvvm.read.ptx.sreg.tid.z` from `int_nvvm_read_ptx_sreg_tid_z` | safe; scope thread; pure; memory none; convergent false | PTX 2.0 on all ([PTX ISA 9.3, 10.1 Special Registers: %tid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-tid)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %tid.z;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::fence::threadfence` | `nvvm.threadfence` | LLVM `llvm.nvvm.membar.gl` from `int_nvvm_membar_gl` | safe; scope device; impure; memory read_write; convergent false | PTX 1.4 on all ([PTX ISA 9.3, 9.7.14.4 Parallel Synchronization and Communication Instructions: membar / fence](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `membar.gl;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::fence::threadfence_block` | `nvvm.threadfence_block` | LLVM `llvm.nvvm.membar.cta` from `int_nvvm_membar_cta` | safe; scope cta; impure; memory read_write; convergent false | PTX 1.4 on all ([PTX ISA 9.3, 9.7.14.4 Parallel Synchronization and Communication Instructions: membar / fence](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `membar.cta;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::fence::threadfence_system` | `nvvm.threadfence_system` | LLVM `llvm.nvvm.membar.sys` from `int_nvvm_membar_sys` | safe; scope system; impure; memory read_write; convergent false | PTX 2.0 on sm_20+ ([PTX ISA 9.3, 9.7.14.4 Parallel Synchronization and Communication Instructions: membar / fence](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-membar-fence)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `membar.sys;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::shared::total_smem_size` | `nvvm.read_ptx_sreg_total_smem_size` | LLVM `llvm.nvvm.read.ptx.sreg.total_smem_size` from `int_nvvm_read_ptx_sreg_total_smem_size` | safe; scope cta; pure; memory none; convergent false | PTX 4.1 on sm_20+ ([PTX ISA 9.3, 10.30 Special Registers: %total_smem_size](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-total-smem-size)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %total_smem_size;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::debug::trap` | `nvvm.trap` | PTX-native `trap` | safe; scope thread; impure; memory none; convergent false | PTX 1.0 on all ([PTX ISA 9.3, 9.7.20.4 Miscellaneous Instructions: trap](https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-trap)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `trap;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::uni_sync` | `nvvm.vote_sync_uni` | LLVM `llvm.nvvm.vote.uni.sync` from `int_nvvm_vote_uni_sync` | unsafe; scope warp; impure; memory inaccessible_read_write; convergent true | PTX 6.0 on sm_30+ ([PTX ISA 9.3, 9.7.14.10 Warp Vote Instructions: vote.sync](https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-vote-sync)) | lowered on `rust-llvm-23.1.3-aab11ee0`; expects `vote.sync.uni.pred <register>, <register>, <register-or-immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::warp::warpid` | `nvvm.read_ptx_sreg_warpid` | LLVM `llvm.nvvm.read.ptx.sreg.warpid` from `int_nvvm_read_ptx_sreg_warpid` | safe; scope cta; impure; memory none; convergent false | PTX 1.3 on all ([PTX ISA 9.3, 10.4 Special Registers: %warpid](https://docs.nvidia.com/cuda/parallel-thread-execution/#special-registers-warpid)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `mov.u32 <register>, %warpid;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::wgmma::wgmma_commit_group` | `nvvm.wgmma_commit_group_sync_aligned` | LLVM `llvm.nvvm.wgmma.commit_group.sync.aligned` from `int_nvvm_wgmma_commit_group_sync_aligned` | unsafe; scope warpgroup; impure; memory read_write; convergent true | PTX 8.0 on sm_90a ([PTX ISA 9.3, Asynchronous Warpgroup Level Matrix Instructions: WGMMA control](https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `wgmma.commit_group.sync.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::wgmma::wgmma_fence` | `nvvm.wgmma_fence_sync_aligned` | LLVM `llvm.nvvm.wgmma.fence.sync.aligned` from `int_nvvm_wgmma_fence_sync_aligned` | unsafe; scope warpgroup; impure; memory read_write; convergent true | PTX 8.0 on sm_90a ([PTX ISA 9.3, Asynchronous Warpgroup Level Matrix Instructions: WGMMA control](https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `wgmma.fence.sync.aligned;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
+| `cuda_intrinsics::wgmma::wgmma_wait_group` | `nvvm.wgmma_wait_group_sync_aligned` | LLVM `llvm.nvvm.wgmma.wait_group.sync.aligned` from `int_nvvm_wgmma_wait_group_sync_aligned` | unsafe; scope warpgroup; impure; memory read_write; convergent true | PTX 8.0 on sm_90a ([PTX ISA 9.3, Asynchronous Warpgroup Level Matrix Instructions: WGMMA control](https://docs.nvidia.com/cuda/parallel-thread-execution/#asynchronous-warpgroup-level-matrix-instructions)) | validated on `rust-llvm-23.1.3-aab11ee0`; expects `wgmma.wait_group.sync.aligned <immediate>;` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`) |
 
 ## Compiler identity and compatibility paths
 
@@ -6751,2167 +6751,2167 @@ This table is generated from the resolved catalog. Evidence stages below disting
 ## Backend-specific lowering evidence
 
 - `abs_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact abs.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fc97be133b128891fbdca9f6a4b9769cfd2712fa35ee81884e52b9b436f6ef32.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbafd197892b5b499ef62752012ccb5e86318641dd9784942872a4ca7d685893. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact abs.bf16x2 route at compute_80 as part of the combined seven-form probe and produced 2,256-byte LTOIR (SHA-256 a9394a9ba0813c36f65bfc2636d7e80b27ab2ffcb6ded14a420d4127a824028d).
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,624-byte sm_80 cubin (SHA-256 1c8ff2eeb9e58f8b3defc22ed929cd48ec2e2f2beab30747f937a9a7869d74d3); cuobjdump showed HFMA2.BF16_V2, HFMA2.BF16_V2.RELU, HMNMX2.BF16_V2, sign/absolute HFMA2 forms, and F2FP.BF16.PACK_AB. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx65`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact abs.f16x2 inline PTX at PTX 6.5/sm_70. An independent typed call to the resolved llvm.nvvm.fabs.v2f16 symbol emitted the same instruction at PTX 6.5/sm_70 and at its native sm_53 floor; sm_52 failed exact intrinsic selection.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the 737-byte PTX module at sm_75 and produced a 3,112-byte cubin (SHA-256 24150b058e7b7ad99ffeacd77588a2f1febdf5479f93f966ac0192c9834babf7). SASS realizes packed absolute value as HADD2 with an absolute source. The PTX 6.4 negative module was rejected because abs.f16/abs.f16x2 requires PTX 6.5 or later. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7282ee206f4fdb3a54d1fe47412d9ab41a11954358a8e0ee203f434988aa5191.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93adc93cea87ea659d8a0b4542f44e1d897cb4c1cd53a06dc08becdac7de7ea6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact abs.f16x2 at compute_75 and produced 1,648-byte LTOIR (SHA-256 e26755d764c7c6576f9bd387d573a868fa247ad1b2c900ad7ab8b8486f1fc79c). The instruction floor is PTX 6.5/sm_53; compute_75 is the CUDA 13.3 profile floor.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,112-byte sm_75 cubin (SHA-256 e780ddaee22735b1c97e35157fae73685685268a445eef4a890d805bb6958c37). cuobjdump shows the same absolute-source HADD2 form as the LLVM/ptxas path. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `active_mask`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.2 / sm_30+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx62`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, side-effect, and inaccessible-memory contract.
-    - backend codegen on `sm_30, ptx62`: `succeeded` — The selected Rust LLVM backend emitted activemask.b32 at the imported instruction floor.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.2 / sm_30+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx62`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a17f68d96e936c632dbeb06ea74eae28b193c05cdfdc08373f0715f348eacf1.
+    - backend codegen on `sm_30, ptx62`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 246910573ff59e4feb1a3b8880eb69d36c7a14f49debfe41f1207db7b4ea51d4.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained exact activemask.b32 PTX and produced 1,812-byte LTOIR from the reviewed probe source (SHA-256 d98ba219074ef5b6b866e18ecff76c8aa6f35d31409cfd4af5b30e07ccd673b8).
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 2,856-byte cubin (SHA-256 d2b3a14122744f1685f4042a7f8653d6bb8d33bf03b540c54bab7da799196557); cuobjdump showed the expected VOTE.ANY active-mask implementation and global result store. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact add.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2c103ab626814fb233a6e15e90a48c2a6a5b78f7f13e38c4c8c6cebca28ef0d0.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af1befea9f33a173ed1a34fa549dfd6e8cf85f450042ec4948cdcd9eef20051f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact add.rn.bf16x2 route at compute_90 as part of the combined three-form probe and produced 1,904-byte LTOIR (SHA-256 87cd4ca7b61dab71aac8885db949e21fbcfbf3bb59ca560c67084d7901ddf60b).
     - device link on `sm_90, ptx93`: `succeeded` — nvJitLink produced a 3,832-byte sm_90 cubin (SHA-256 b0715cda56457dd1b90115ad52c73c8bb920df6f0f4e08df60410cfb4a57635a); cuobjdump showed HFMA2.MMA.BF16_V2 for add, HADD2.BF16_V2 with a negated RHS for sub, and HMUL2.BF16_V2 for mul. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b7f49dda4585f12ddfc9ecaf71517120de1aefd78a9f843fcb511f81bbabaf4.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b33cc169935b856e4ff58245496b889d56c48c772653634f963facb395c24326. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_75 as part of the combined four-form probe and produced 1,932-byte LTOIR (SHA-256 edb864f3999af5c06558f9dde87cd506ac0c2188cb6a71b19fbf8b33d6b5e5ae). The instruction floor is PTX 4.2/sm_53; compute_75 is this libNVVM profile's floor, and compute_70 is rejected as an unsupported profile option.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,368-byte sm_75 cubin (SHA-256 5ca3d1ca1d9c56802bb78bdad8f5e1138b17dc4dc9f96a7b1e9572eb4010d50c). cuobjdump SASS contains HFMA2, HADD2 add/sub forms, and HMUL2, matching the LLVM/ptxas path. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `add_ftz_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `add_rm_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rm_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rm_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rm_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rm_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rn_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rn_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rn_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rn_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rn_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rp_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rp_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rp_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rp_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rp_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rz_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 20 candidate declarations. The 12 selected typed routes matched their imported signatures and property sets. The canonical manifest bundle has SHA-256 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60; a physical audit matched all 100 payloads in the artifact index with SHA-256 7523944499f326a8eb1090c4c82a5efe3b7ff319dd806d997b814508d61dff3a.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted every direct add form exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 5dcbdc5af2a1d246222e549b948fdf525f75951ea78877c7ef2b1f78bffc4842. No subtraction or mixed f16/bf16 selection was admitted.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all 20 fresh sm_80/PTX 7.0 outputs. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dd11e1e9f47084b276eb70bc5ad65ca42d3c1a5f357e490477b10e470b5ae8d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b17c7a77acdeaccbb6ffd876b1efe91d4507ecb15d7e9a2515c04f96dc44807.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0736508303112b799ebabcdcfa600ecfdef2a3c9ea9548cd80e1bb5316831c64. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rz_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `add_rz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 4,719-byte typed-pointer NVVM IR fixture (SHA-256 4115a5696631ec940f8f90cb4f70c21f929f58fa259b34533528e58b11c0968b) to 3,023-byte PTX 9.3 (SHA-256 943adbcba0c1004fe6d4bff9af24f15c965ab426c1b47b878a9425a70ce5bb5f) and 3,540-byte LTOIR (SHA-256 1b10b69dae868b3f058b31d75e855d5433966b87af138cf74cf3b8a99af46e3c). The PTX retained all 20 exact, unique add forms with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 4,008-byte sm_80 cubin with SHA-256 bf08ec6f01a1a355c9d39c967d133ff8146de0cfb5da01bb5f1679cbc0551b27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 4,008-byte cubin with SHA-256 55b093a238d90be57b1962928326d1ca597ad54189fa2097b3e2207177d708c0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cluster_arrive`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-barrier` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — libNVVM accepted source IR SHA-256 311879881b865fd8094b574f4e4cef82be548f2349eef07d8abffb5f9608870c but emitted six unresolved extern functions instead of barrier.cluster instructions; LTOIR SHA-256 5c8e1b263481c6281022e9bdd0d73a870abfd8ec4e849163bb40abc91e241b2f and PTX SHA-256 81a5de0a13d9f0ebba19e108b9ea74cba3ae0f3e2efaf8561dd7f1f1f279cb18.
     - device link on `sm_90, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after rejecting the generated PTX near the unresolved .extern function names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,092-byte LTOIR with SHA-256 741b16f401ce6dbe85c24eaaf48eb3b49eca46336ae02cf152dce734d83f0151 and PTX SHA-256 6c7b50ae9f833d4bab127cb83ba22571cdf544c63b0b993399122b52d8ba8a7e.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 4,200-byte cubin with SHA-256 bd103d9213745d04e97d3b046b154c95329b03ddbd1d75ea5e04d9bb3e852714; cuobjdump retained four UCGABAR_ARV and two UCGABAR_WAIT instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cluster_arrive_aligned`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-barrier` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — libNVVM accepted source IR SHA-256 311879881b865fd8094b574f4e4cef82be548f2349eef07d8abffb5f9608870c but emitted six unresolved extern functions instead of barrier.cluster instructions; LTOIR SHA-256 5c8e1b263481c6281022e9bdd0d73a870abfd8ec4e849163bb40abc91e241b2f and PTX SHA-256 81a5de0a13d9f0ebba19e108b9ea74cba3ae0f3e2efaf8561dd7f1f1f279cb18.
     - device link on `sm_90, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after rejecting the generated PTX near the unresolved .extern function names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,092-byte LTOIR with SHA-256 741b16f401ce6dbe85c24eaaf48eb3b49eca46336ae02cf152dce734d83f0151 and PTX SHA-256 6c7b50ae9f833d4bab127cb83ba22571cdf544c63b0b993399122b52d8ba8a7e.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 4,200-byte cubin with SHA-256 bd103d9213745d04e97d3b046b154c95329b03ddbd1d75ea5e04d9bb3e852714; cuobjdump retained four UCGABAR_ARV and two UCGABAR_WAIT instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cluster_arrive_relaxed`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the convergent nocallback declarations; source IR SHA-256 2e2c065b1abe60a62a61678dca92352084f523640ebf0feb0dd47e0f8417b18f and bitcode SHA-256 7522f3d31dd29407d8c5d8b4c35ad5f38f663067ad3c415effb5d096b4d93463.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — LLVM emitted each exact instruction once; PTX SHA-256 4e8650939b44a80b88fe91ede1a9568ac5526889254c9d6954d52176c10a9714.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 698cf65618fd422030be4645284d6ab56b24f75d0e955a49380ef78bb8e044a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c5b7ef9743a0c36a3ad27cf3093c5d3160111a0acee880922a93dab4ce330a7.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 462bdc8a89623f986fa9c307a069d444805dcb8644ef7f9793f45a48a5cc94c6.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5501565e9a498f7f5dba887004bc2cf506c57249b8a4a2e8fdebcefc8007b146. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-barrier` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — libNVVM accepted source IR SHA-256 311879881b865fd8094b574f4e4cef82be548f2349eef07d8abffb5f9608870c but emitted six unresolved extern functions instead of barrier.cluster instructions; LTOIR SHA-256 5c8e1b263481c6281022e9bdd0d73a870abfd8ec4e849163bb40abc91e241b2f and PTX SHA-256 81a5de0a13d9f0ebba19e108b9ea74cba3ae0f3e2efaf8561dd7f1f1f279cb18.
     - device link on `sm_90, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after rejecting the generated PTX near the unresolved .extern function names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,092-byte LTOIR with SHA-256 741b16f401ce6dbe85c24eaaf48eb3b49eca46336ae02cf152dce734d83f0151 and PTX SHA-256 6c7b50ae9f833d4bab127cb83ba22571cdf544c63b0b993399122b52d8ba8a7e.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 4,200-byte cubin with SHA-256 bd103d9213745d04e97d3b046b154c95329b03ddbd1d75ea5e04d9bb3e852714; cuobjdump retained four UCGABAR_ARV and two UCGABAR_WAIT instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cluster_arrive_relaxed_aligned`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the convergent nocallback declarations; source IR SHA-256 2e2c065b1abe60a62a61678dca92352084f523640ebf0feb0dd47e0f8417b18f and bitcode SHA-256 7522f3d31dd29407d8c5d8b4c35ad5f38f663067ad3c415effb5d096b4d93463.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — LLVM emitted each exact instruction once; PTX SHA-256 4e8650939b44a80b88fe91ede1a9568ac5526889254c9d6954d52176c10a9714.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 698cf65618fd422030be4645284d6ab56b24f75d0e955a49380ef78bb8e044a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c5b7ef9743a0c36a3ad27cf3093c5d3160111a0acee880922a93dab4ce330a7.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 462bdc8a89623f986fa9c307a069d444805dcb8644ef7f9793f45a48a5cc94c6.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5501565e9a498f7f5dba887004bc2cf506c57249b8a4a2e8fdebcefc8007b146. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-barrier` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — libNVVM accepted source IR SHA-256 311879881b865fd8094b574f4e4cef82be548f2349eef07d8abffb5f9608870c but emitted six unresolved extern functions instead of barrier.cluster instructions; LTOIR SHA-256 5c8e1b263481c6281022e9bdd0d73a870abfd8ec4e849163bb40abc91e241b2f and PTX SHA-256 81a5de0a13d9f0ebba19e108b9ea74cba3ae0f3e2efaf8561dd7f1f1f279cb18.
     - device link on `sm_90, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after rejecting the generated PTX near the unresolved .extern function names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,092-byte LTOIR with SHA-256 741b16f401ce6dbe85c24eaaf48eb3b49eca46336ae02cf152dce734d83f0151 and PTX SHA-256 6c7b50ae9f833d4bab127cb83ba22571cdf544c63b0b993399122b52d8ba8a7e.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 4,200-byte cubin with SHA-256 bd103d9213745d04e97d3b046b154c95329b03ddbd1d75ea5e04d9bb3e852714; cuobjdump retained four UCGABAR_ARV and two UCGABAR_WAIT instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cluster_wait`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-barrier` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — libNVVM accepted source IR SHA-256 311879881b865fd8094b574f4e4cef82be548f2349eef07d8abffb5f9608870c but emitted six unresolved extern functions instead of barrier.cluster instructions; LTOIR SHA-256 5c8e1b263481c6281022e9bdd0d73a870abfd8ec4e849163bb40abc91e241b2f and PTX SHA-256 81a5de0a13d9f0ebba19e108b9ea74cba3ae0f3e2efaf8561dd7f1f1f279cb18.
     - device link on `sm_90, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after rejecting the generated PTX near the unresolved .extern function names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,092-byte LTOIR with SHA-256 741b16f401ce6dbe85c24eaaf48eb3b49eca46336ae02cf152dce734d83f0151 and PTX SHA-256 6c7b50ae9f833d4bab127cb83ba22571cdf544c63b0b993399122b52d8ba8a7e.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 4,200-byte cubin with SHA-256 bd103d9213745d04e97d3b046b154c95329b03ddbd1d75ea5e04d9bb3e852714; cuobjdump retained four UCGABAR_ARV and two UCGABAR_WAIT instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cluster_wait_aligned`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the convergent nocallback declarations; source IR SHA-256 7e45b48ac5a4963ffc711c20008807fa871f47df35455be3a89e60d6a2f96717 and bitcode SHA-256 7fc4b04573811961a3141aed2aca46f436d38c45dc82d0a7859f64294cfef10c.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — LLVM emitted each exact instruction once; PTX SHA-256 c567819a2ce45e02fbf006d66232f507bad410d93ed30eed81c4917b580dd909.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 188b14eec5d708849572bfa258ae17df6293d02bb224c4638294af18d1a68865. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 709ac5ee2e1dd7825ecc0de9afa6ac275df2df2edeadc6775491baa136592e22.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e1e88cd16c49e09653b14a27a2bc796381aa018d17067a47b7983db21ae09f3.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d7c3f4ea8fcd98c2c86bed72381ab990ad0e38ec3f9ca467a4dfc2fae748b37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-barrier` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — libNVVM accepted source IR SHA-256 311879881b865fd8094b574f4e4cef82be548f2349eef07d8abffb5f9608870c but emitted six unresolved extern functions instead of barrier.cluster instructions; LTOIR SHA-256 5c8e1b263481c6281022e9bdd0d73a870abfd8ec4e849163bb40abc91e241b2f and PTX SHA-256 81a5de0a13d9f0ebba19e108b9ea74cba3ae0f3e2efaf8561dd7f1f1f279cb18.
     - device link on `sm_90, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after rejecting the generated PTX near the unresolved .extern function names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,092-byte LTOIR with SHA-256 741b16f401ce6dbe85c24eaaf48eb3b49eca46336ae02cf152dce734d83f0151 and PTX SHA-256 6c7b50ae9f833d4bab127cb83ba22571cdf544c63b0b993399122b52d8ba8a7e.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 4,200-byte cubin with SHA-256 bd103d9213745d04e97d3b046b154c95329b03ddbd1d75ea5e04d9bb3e852714; cuobjdump retained four UCGABAR_ARV and two UCGABAR_WAIT instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `barrier_cta_arrive`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 6844935a6b2fc799ecf3245780589f5de02583685caae0267be1f37668535a2b, bitcode SHA-256 d014888592c0d71d5c569711143349e9daaea3079578144385a41fa055e87992, and canonical IR SHA-256 e2a2a5029602f80fb55f6f90e96083d51ce76da3c019b666aad025d0e64c0987.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — LLVM emitted all four reviewed operand forms in PTX SHA-256 94113e81da1085d1688c37f811dc497e0963be3af0f703f03862268e4a14678f.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 70c603a8e3092012e74c6e6f5a2330489176c4af38499c379753f095ae899a2a.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6a153e4c537c56cc43402bcf7786c6540c696ff807919d11973d650c7bc9a78c.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9499277c4334862a892ac0c79eb49c79fc5c0dc889639a4b22dd5d0ea2134d0c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 6.0 / sm_75+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 65671f4a746d155eb12254e1f871fcd3fead7041b26f32c06a9ce819bbc90d3b to LTOIR SHA-256 147b4026da4581071480f95cf1f691b4888078a572d050640bc71bedba935ed4; nvJitLink emitted matching PTX SHA-256 2a9d61847160b457cd8dffbcb0ff18d6bfaffc0bc5d45175dc4515dfce786a77.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced cubin SHA-256 1fd529cda3a3e0acf3310eec5e68b025d6b41d6e12588f45cdf1d9fbf488d1cf; cuobjdump retained BAR.ARV. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `barrier_cta_arrive_aligned`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 042ab4e18a0df19c17f139123fd21759a9b179221f8d303f9d13c8079484f5cf, bitcode SHA-256 8c4a5ec77772763a7deb91899e5ae3a1fd10df067d619b985b4e6e971e8b2319, and canonical IR SHA-256 6ce3356d92a9740dfeb0f53cf4aa551586a1ee201a2210dadf631131e26ce047.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all four reviewed operand forms in PTX SHA-256 6b4d17807a12f8298fca5f4957b8e3da5ed27fcbd47111f8a173bbe61d5c8aad.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 171894b0dd863526a8ff4cf86ecc05f04a47032b49254ae9d5a03f58bcc5eaf4.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cb96f04a02a2ae0cde2c1d3691ca3d2752577a6b32337b39ab1afecaa7f98249.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11118e20cc0d2ec90152a19b6e75a9dbcc4f09a1d5698c030ebf890c88095fa1. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 1.0 / sm_75+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 6f4a98ea80bfcad48693cdfd222167642dbc916957e2f2dd74bcd5b2ee8550e8 to LTOIR SHA-256 60404c53823ad015a8a1ea47c3bb8d705d833f6bd8184b569933c17e74d98710; nvJitLink emitted matching PTX SHA-256 130dc6b1c3ac18204ed037ec7ada696f28829255d62ff2f3971c32bdbdec86af.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced cubin SHA-256 a73a41d6b67c452bd0b5d73dcb28e01a81c01cc91d27a6e98f499a409d4f6001; cuobjdump retained BAR.ARV. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `barrier_cta_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 73bcf70a8647d141dea6d68aba4ab626818aef73c25eaf85b9c415814dcf0e71, bitcode SHA-256 50b842e412973300a1b0c67190c173f7f5f744db15efc9d639daa3b5c7ec33c6, and canonical IR SHA-256 d068e7a4ff32510dd439f5e92e97374a319e4ba6c1a6380cff5c855a0cde6a3f.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — LLVM emitted all four reviewed operand forms in PTX SHA-256 b220ca5fdae82f6048cc1e930fa07a9f6e8cf23a610476f0cf469a01e985a85d.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 ecde7ac1ed76dafe802f081e6dfa4a3362c265aff18a66e49eda008ee772e857. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1ec32a6b202b53ac34f270d293fd60d90d5d6de4dacef2b7d7912ab16698c69f.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa2ed67b2503e249a9ae63ef2a7436c41d098c27956fe0faf441c0e961db4151.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a9e71fa97f369891143a3c32fa8c80f1c28be09176445e1edfff81c54615ffd. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 6.0 / sm_75+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 7c4289764865653725853dc986143d5dde7ef39f05f597aa54ad99d44e007d79 to LTOIR SHA-256 f31807cefa0865b1f9e3284dd06b2d3d5718913fa37a2618c57dfec4b11cdb52; nvJitLink emitted matching PTX SHA-256 fe2a971c69e1bab0ab620a3d6f97e55cbb6b36f46a92743989a586b58f988889.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced cubin SHA-256 24b9e4d4395e926a4aa66ccfcc439ad28955798825178c7113076b1ccd579d0e; cuobjdump retained BAR.SYNC. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `barrier_cta_sync_aligned`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 f5faf9f6ac1de5de821476c35d6cfed39ffb9abc1de2e7ac9038c6f51c7bda6c, bitcode SHA-256 3c90f0f9686014d1074c753f644ca9dac74397f465cb78f23e726820af47345d, and canonical IR SHA-256 70ee92254f4aa29db9731a3beddef77865f6f5bff7a81a7c66aa076726076ea4.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all four reviewed operand forms in PTX SHA-256 d3fbf9d7aa025ee2f4836094b8652fb29a5557e94bbc73689c5b18dd04edc47a.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 64228b359e064593cb479b86c6c493fea7ed182e88586a2066cbc1baf81a3b05. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c313db28069767d98c8d477dd5512e1d4dae29dc52ea0d691fe03edb03c5e50.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0615f66795bc5a379d6cf2426d83a39d4baae46194d61b6cde902e9b8c8ab367.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 999e27bbd3d4ccd52c8f1366731b70ab35d9d67ed19125ffb66ac7bf160b4bbc. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 1.0 / sm_75+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 579466b9aa311ff37495c4e7f604bdf107ae171d820b25105feb84258152f7f0 to LTOIR SHA-256 94e983ade64421ea5a8d44d8453affd15dc765762667343044f76efea8e1aa16; nvJitLink emitted matching PTX SHA-256 6b10b5c6a1248b83b5f4518bdbb8876e15730cc1b7b2b2796da142ece753b0ef.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced cubin SHA-256 2ed2b6758f44f5bfe3414a3f0051f915709eddd6b4054401ec85aa3afbcc942f; cuobjdump retained BAR.SYNC. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `breakpoint`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-debug-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-debug-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-debug-control` at PTX 3.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 emitted a 562-byte PTX 9.3 module containing all three instructions; source SHA-256 2e8b8ffd062afdcf1c1806d526388b9bacc7469918370d26af9186b2c9a4b1e1 and PTX SHA-256 ba2d42ec37ae06d4a460291a966872cbc35e999ffd0aa72c7e1aa6cf0f54e6f8.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 produced a 5,416-byte cubin with SHA-256 d9c0af472531aa6da4a84efc0b4cbaccc56f9e090c5c3160792842172c225ee1; cuobjdump retained PMTRIG and BPT.TRAP instructions. Tool `/usr/local/cuda-13.3/bin/nvcc` reports `CUDA compilation tools, release 13.3, V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `6b55d91cd23f314ccb6801d71d64d20760a7ec31d37d87017ea0681d79bd3285`). Artifact `cubin`.
 - `clc_query_get_first_ctaid_x`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100`: `succeeded` — llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-clc` at PTX 8.6 / sm_100+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_100`: `succeeded` — CUDA 13.3 libNVVM verified the 1,898-byte source with SHA-256 37ddba55e2ee9cd6643285442a3eb625a2a3cb626d8da8627ab0b7ccf9910b15.
     - backend codegen on `sm_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 3,096-byte LTOIR with SHA-256 ef38dfe4cca9b5ef09d0ba5d0eefa3874c24ef29af751f880a461e1567f28db4. nvJitLink's 1,700-byte PTX had SHA-256 dd4e2baaeb1704866cbe790baf7880a7f71d17b8f6231b9d3a778a47453cb4ee and retained all five exact instructions with no extern calls.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,000-byte cubin with SHA-256 3d1ae4180befb13753735b0457f9292ec10d593873dee4e77ef1fb634dfa3aea. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clc_query_get_first_ctaid_y`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100`: `succeeded` — llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-clc` at PTX 8.6 / sm_100+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_100`: `succeeded` — CUDA 13.3 libNVVM verified the 1,898-byte source with SHA-256 37ddba55e2ee9cd6643285442a3eb625a2a3cb626d8da8627ab0b7ccf9910b15.
     - backend codegen on `sm_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 3,096-byte LTOIR with SHA-256 ef38dfe4cca9b5ef09d0ba5d0eefa3874c24ef29af751f880a461e1567f28db4. nvJitLink's 1,700-byte PTX had SHA-256 dd4e2baaeb1704866cbe790baf7880a7f71d17b8f6231b9d3a778a47453cb4ee and retained all five exact instructions with no extern calls.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,000-byte cubin with SHA-256 3d1ae4180befb13753735b0457f9292ec10d593873dee4e77ef1fb634dfa3aea. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clc_query_get_first_ctaid_z`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100`: `succeeded` — llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-clc` at PTX 8.6 / sm_100+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_100`: `succeeded` — CUDA 13.3 libNVVM verified the 1,898-byte source with SHA-256 37ddba55e2ee9cd6643285442a3eb625a2a3cb626d8da8627ab0b7ccf9910b15.
     - backend codegen on `sm_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 3,096-byte LTOIR with SHA-256 ef38dfe4cca9b5ef09d0ba5d0eefa3874c24ef29af751f880a461e1567f28db4. nvJitLink's 1,700-byte PTX had SHA-256 dd4e2baaeb1704866cbe790baf7880a7f71d17b8f6231b9d3a778a47453cb4ee and retained all five exact instructions with no extern calls.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,000-byte cubin with SHA-256 3d1ae4180befb13753735b0457f9292ec10d593873dee4e77ef1fb634dfa3aea. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clc_query_is_canceled`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100`: `succeeded` — llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-clc` at PTX 8.6 / sm_100+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_100`: `succeeded` — CUDA 13.3 libNVVM verified the 1,898-byte source with SHA-256 37ddba55e2ee9cd6643285442a3eb625a2a3cb626d8da8627ab0b7ccf9910b15.
     - backend codegen on `sm_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 3,096-byte LTOIR with SHA-256 ef38dfe4cca9b5ef09d0ba5d0eefa3874c24ef29af751f880a461e1567f28db4. nvJitLink's 1,700-byte PTX had SHA-256 dd4e2baaeb1704866cbe790baf7880a7f71d17b8f6231b9d3a778a47453cb4ee and retained all five exact instructions with no extern calls.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,000-byte cubin with SHA-256 3d1ae4180befb13753735b0457f9292ec10d593873dee4e77ef1fb634dfa3aea. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clc_try_cancel`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100`: `succeeded` — llvm-as accepted the 1,394-byte source with SHA-256 895fe9619dfbb6be46fc38eec6731ec5e5fee6db392b2033b2b8f246a5e65a5f and produced 2,412-byte bitcode with SHA-256 fdd42b00f2b88733727f997b2d8527911593601acc2d49d2379c1af23735528b.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — The pinned Rust LLVM emitted all five exact CLC instructions with no extern calls. The 1,651-byte PTX had SHA-256 6f0d5158aa6786a127524302e620110064063991155aea99ce634290803b6996.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas produced a 6,008-byte cubin with SHA-256 257a5c67bb688844d2dec63c2351d13b987918a5b6e8c324558be33d99e3b15d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-clc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49b5ca46ca7458a71b6b1c34447f73fdde78ac1e1a04e7e9bf8aff334b1e006e.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 90c7f9c7bf8b3c848039f3b742865eb24507dcfd95f3996a29bebf9fd2ce1316.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0f08792c114374ae43fc3924ba7a3e5197102950e2ba89280406d086240119e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-clc` at PTX 8.6 / sm_100+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_100`: `succeeded` — CUDA 13.3 libNVVM verified the 1,898-byte source with SHA-256 37ddba55e2ee9cd6643285442a3eb625a2a3cb626d8da8627ab0b7ccf9910b15.
     - backend codegen on `sm_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 3,096-byte LTOIR with SHA-256 ef38dfe4cca9b5ef09d0ba5d0eefa3874c24ef29af751f880a461e1567f28db4. nvJitLink's 1,700-byte PTX had SHA-256 dd4e2baaeb1704866cbe790baf7880a7f71d17b8f6231b9d3a778a47453cb4ee and retained all five exact instructions with no extern calls.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,000-byte cubin with SHA-256 3d1ae4180befb13753735b0457f9292ec10d593873dee4e77ef1fb634dfa3aea. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clc_try_cancel_multicast`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-clc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a or sm_120a or sm_121a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a`: `succeeded` — llvm-as accepted the 390-byte source with SHA-256 e4ed73cd8c39af86e8723ca03126c44f31cc9511a73b6ece710b7875bc128b14 and produced 1,912-byte bitcode with SHA-256 2d37df5745e6221f4b75791540de98edab835e3370f36043bcf6aa7af4a3c22d.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — LLVM 22 emitted one exact multicast instruction with no extern call for sm_100a/PTX 8.6, sm_101a/PTX 8.6, sm_103a/PTX 8.8, sm_110a/PTX 9.0, sm_120a/PTX 8.7, and sm_121a/PTX 9.0. Their PTX SHA-256 values were 366b5155464356e2e01d29fb3ad209f365ba68fc2e540489ce7e2f5830658cde, bf925481f00d8023b5493cf2daee655ba90e32486d16b140c013df87bc1b5625, 6310890cf30d93ae0d02d63362350bfd13428ec4522358d80b459a3b1e86435b, 7ae967df1894e5a0158eabed03a607eeb971d1d2f7e5ad2892ca7f632d95ccd7, a4d5dd50691a5e0d3a32d4c77bf87258009ecebf2d66d00bb73292d0cbb4fa4e, and c52b501ba903c302697c994c9ea3abd4506b8a3aee1fd6ca320cad533140c894. Plain sm_100 failed instruction selection, as required.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas assembled the exact probes for sm_100a, sm_103a, sm_110a, sm_120a, and sm_121a. Their cubin SHA-256 values were 30160dcbe4de4c64e9b78f907e2663cdfb244fbe1a5dfbba66a6b8a1c52d5d76, 357d54085a3c661f8b0841d933d2d1600a688b54a5004ac0610469c35d016c57, 6d3afe7f862c5f2b5931972aa5c5dac9e2649f14407dfe67ade547b72af951db, e481feac9e4382025b1204c38614a1f5bdf141f9ce226da45d4ff835d58fdaba, and dcb515f3f0bb87fb6ffb2e74b78542979a0264f56b6df43a3913786e201c44f9. CUDA 13.3 does not expose the LLVM-only sm_101a name. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-clc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a or sm_120a or sm_121a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd29e45eadad739b58e43893296b4b8ac5e05d43c3c3298836fb126503d61e83.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 423445d4b9994ee20be92cbabc0ad7c8ea667c6360848df3b2a5be7020df4051.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f8543fdf61a5a323f880af2fdfe1b86d2ee1311a024f5a9e88d6c40a23845925. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-clc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a or sm_120a or sm_121a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_100a`: `succeeded` — CUDA 13.3 libNVVM verified the 789-byte source with SHA-256 41969303d126b015e653137ac3ebf7cca283544cd0f2a85067a631241ee27582.
     - backend codegen on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,484-byte LTOIR with SHA-256 e18003b9fd91884b03a64dcabad7f091061d9a28300346ff815b6b5499e8b6bd. nvJitLink's 677-byte PTX had SHA-256 6d65ac91ef71804809bb05a81d454b1598a83df7ba1f1bc45f4635e582419678 and retained the exact multicast instruction with no extern call. Plain sm_100 failed terminal compilation, as required.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 5,408-byte cubin with SHA-256 16e900a28f60744304dc5629f4742d971dbc5b7438cb895f5bbfd2dea2ea031f. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clock`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 1.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75`: `succeeded` — CUDA 13.3 libNVVM verified all six intrinsic declarations before compilation.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,316-byte LTOIR (SHA-256 57b3175e40f2a217ede37a7b8dc3aadce172260ec9ed5459eca6b2d7023711bb) and exact 879-byte PTX (SHA-256 dfac22a9564622fa9b2e9ef64f97a1a50e0e23c853f9734e2aa7c5b8a22ad8eb).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 276fc6a71465f9db688bafa30cd6f85084f291f9fab7a7155e090aa3db1173d3; cuobjdump retained clock and device-constant register reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `clock64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75`: `succeeded` — CUDA 13.3 libNVVM verified all six intrinsic declarations before compilation.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,316-byte LTOIR (SHA-256 57b3175e40f2a217ede37a7b8dc3aadce172260ec9ed5459eca6b2d7023711bb) and exact 879-byte PTX (SHA-256 dfac22a9564622fa9b2e9ef64f97a1a50e0e23c853f9734e2aa7c5b8a22ad8eb).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 276fc6a71465f9db688bafa30cd6f85084f291f9fab7a7155e090aa3db1173d3; cuobjdump retained clock and device-constant register reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cos_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cos_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_commit_group`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_1d_l2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_1d_l2_cache_hint`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_2d_l2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_2d_l2_cache_hint`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_3d_l2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_3d_l2_cache_hint`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_4d_l2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_4d_l2_cache_hint`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_5d_l2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_5d_l2_cache_hint`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_gather4_2d_l2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a`: `succeeded` — llvm-as and llvm-dis canonicalized both declarations. SHA-256 over sorted per-operation ID/hash lists is 4c60ef82171ca5e9bdd92a90683367bbb73023593d27e1af329fa9af9cbce511 for source IR, d186a1e64a5832f4bbe70323a18796cfaef6ab5d4b685b32709085ff1c24e960 for bitcode, and 609578ee9ee7d3bc1bb280cdf57aaa6b7bb1ea1a341a159ff3dec47c0d235ec0 for canonical IR.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The pinned LLVM emitted both exact instructions at their catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 d5149bf319d478db6e1e40978c5acb62203fbb92fa5d767f0a4ddeaa3eb8fb8e.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted both modules; the sorted per-operation ID/cubin-hash list has SHA-256 3a914c3e352965013e8a3b1a937f4faa4b4966830f04ca63a3c4d982427cf3d5. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcf9f4e3b994a7ccfade3699fa2c7393855b07aee83a70d3c449d4f4260098c4.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 701c7338ba48b42868907f3900d1074b019fab8e257f5676f9a92e368c4718f1.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0635d30976c9afa0d905c53f55599ce0fa36a009828f00a5ae0f5283857a21de. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_100a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained both exact instructions. SHA-256 over sorted per-operation ID/hash lists is 106b4563b36b19517cb93a490eec8e614b0287952853f048bd4f9ab51f4ab7b6 for source IR, 67725376d1aaa95a86251a28b2158603b25a066847a56b78dbef90737af753a8 for LTOIR, and e869be13d42d49b7c1569fcdf208b58f5009d36c99c8f6658680b39554e61b43 for linked PTX.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 423d18d77ff708bea6b9f7d19f9628ceff7b0e12dae651b012889ac19a86bc5a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a`: `succeeded` — llvm-as and llvm-dis canonicalized both declarations. SHA-256 over sorted per-operation ID/hash lists is 4c60ef82171ca5e9bdd92a90683367bbb73023593d27e1af329fa9af9cbce511 for source IR, d186a1e64a5832f4bbe70323a18796cfaef6ab5d4b685b32709085ff1c24e960 for bitcode, and 609578ee9ee7d3bc1bb280cdf57aaa6b7bb1ea1a341a159ff3dec47c0d235ec0 for canonical IR.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The pinned LLVM emitted both exact instructions at their catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 d5149bf319d478db6e1e40978c5acb62203fbb92fa5d767f0a4ddeaa3eb8fb8e.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted both modules; the sorted per-operation ID/cubin-hash list has SHA-256 3a914c3e352965013e8a3b1a937f4faa4b4966830f04ca63a3c4d982427cf3d5. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcf9f4e3b994a7ccfade3699fa2c7393855b07aee83a70d3c449d4f4260098c4.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 701c7338ba48b42868907f3900d1074b019fab8e257f5676f9a92e368c4718f1.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0635d30976c9afa0d905c53f55599ce0fa36a009828f00a5ae0f5283857a21de. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_100a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained both exact instructions. SHA-256 over sorted per-operation ID/hash lists is 106b4563b36b19517cb93a490eec8e614b0287952853f048bd4f9ab51f4ab7b6 for source IR, 67725376d1aaa95a86251a28b2158603b25a066847a56b78dbef90737af753a8 for LTOIR, and e869be13d42d49b7c1569fcdf208b58f5009d36c99c8f6658680b39554e61b43 for linked PTX.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 423d18d77ff708bea6b9f7d19f9628ceff7b0e12dae651b012889ac19a86bc5a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_1d_g2s`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_1d_s2g`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_2d_g2s`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_2d_g2s_multicast`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_2d_g2s_multicast_cg2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 521-byte source had SHA-256 c8f988c762ae488faa6a7a31f1391a74b33918a088449f4ec2bfbdb1ca33a201; the 1,964-byte bitcode had SHA-256 dd23f8a1a678731e95b7e2283da19530bce758b7b55ded0c98d01e99c971b978.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The pinned LLVM emitted the exact cta_group::2 instruction with no extern call. The 1,098-byte PTX hashes were 88d30f75f80d7603ec5852572942e3406db97d10a49bdb7b65c64fa1bcea2257 for sm_100a, ed24ee1e914179959d69b7b153d5d65bbead0c864bfeb624871f9c19f5cc17f3 for sm_101a, 451331c3b63021119d0e9ccdc7617e70b505943c69e1066d193ef1b74b54523e for sm_103a, and 13c7be98f933799fafa4055423395bc1d5d4c2f4303e70f27c38ded9a0e1ace8 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubins for sm_100a, sm_103a, and sm_110a with SHA-256 7c522a3420f9856b526e23c6e3dc4067d244b27095057b91ed1b4af27a8abbcf, b3e3934adc407c847b428d93936f7d8e170059f4004218774c8755f71d599419, and ef654bdc632b1887f8622e765f8f5a2d0571b97c28eb39f17fd355483d8b9d31. Separate probes rejected sm_120a and sm_121a because cta_group::2 is unsupported. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 065ffbad7cc3f4a868f3de9e7f86d3a4fb1b0ea7047e1ddcdd578192e1fe101c.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6f52db1ccb491246b76348fb43a73fc961823d087d80432a2cec08dc36439f1.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eea1fea3e3abbce7a6d377fafd8ad8e2c8fec998a990afbd664b7ce4f3856dff. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained the exact cta_group::2 instruction with no extern call. The source IR SHA-256 is be32e3ad9cf01c97078b9a6799a994afc5e1c1264fd0ad0831b2e98f6f2275ca, LTOIR SHA-256 is ebc556f17c97cd4da6eb87512373c3d93bd44ba3faf68db41206f934a22a60ac, and linked PTX SHA-256 is 8b48f700abf6f55e3dc13443bf1ffe7a45f639cd16f683443cfbf5c84feacc38.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a valid sm_100a CUDA ELF cubin with SHA-256 07a9b441996fc5ba0b8a05aebf0ce67218f0ea8e75f45e4c652a0778207a5343. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_2d_s2g`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_3d_g2s`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_3d_s2g`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_4d_g2s`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_4d_s2g`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_5d_g2s`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_5d_s2g`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_add_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_and_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_dec_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_inc_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_max_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_min_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_or_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_im2col_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_im2col_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_im2col_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_tile_1d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_tile_2d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_tile_3d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_tile_4d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_tensor_reduce_xor_tile_5d`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized all 64 reduction modules. SHA-256 over the sorted intrinsic-ID/source-hash manifest is f85d0bc10b19d1140332a33acff2cc75120f16b3e0088e74378c619a48723ec7 for source IR and a95c17c7b963afa869a9304312f100e36b7ecce139840d8456c74834d98398c6 for bitcode.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 64 exact cp.reduce.async.bulk.tensor instructions with no unresolved LLVM intrinsic calls. SHA-256 over the sorted intrinsic-ID/PTX-hash manifest is 71393a42e9bf1a7620f0e242119d91e38fd667b28634cba6d5d4a02478d70877.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3.73 ptxas assembled all 64 modules. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is 2c553dc603c4efbef0ee26812c373f238d4c1bb46507d77473ecd0831b42a920. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.73 build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e6c07a1e38ba5d7919b12f3d3286addbc62387ea95e27bec48769dfff546e66.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a716e70847f67f8313d9b79a1625486bb1a0690df021f2191feda1e4b16277cc.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 64 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8e767deaa1398c0c43700dd0a91d40f67b8c2ea19f4211ee206a2129590c84e9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma-reduce` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM accepted all 64 LegacyLlvm7 modules (NVVM IR 2.0, debug IR 3.1) and retained every exact cp.reduce.async.bulk.tensor instruction through LTO, including im2col_no_offs, with no extern calls. SHA-256 over the sorted intrinsic-ID/hash manifests is c06e4126517f7bee2e461132e9567bb0d0b630210f3f996950fe9ffdba6b6c5e for source IR, c3823a7bdf4743cf978c00f573b49fc727b109d2fa914e95fbd7f8475d3beb3e for LTOIR, and 9ee7bfc599192e463627b143fe03c93b4b8a77aa53d6d3361b0d502f305ea37b for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins for all 64 variants. SHA-256 over the sorted intrinsic-ID/cubin-hash manifest is dd89ca7b9557e7743fe946140abcdcfd590bbe922312284e18fc977ef1f389b0. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_bulk_wait_group`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_bulk_wait_group_read`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as canonicalized the intrinsic attributes. The 3,580-byte source had SHA-256 c87e60df3991da94c6997bfd9bb14b177031d3cf77567bebb569542c06866929; the 3,116-byte bitcode had SHA-256 00ce86db99146a468482d5471cdbb30bf092703bdb8e73d975b473e3e8db4d83.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted all 14 exact instructions with no extern calls. The 2,495-byte PTX had SHA-256 451ccb2843bfc650cf19991aaed55bcb69468de996534cdfbdca5b3f7e83916a.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 1,720-byte cubin with SHA-256 ea137713143ecca0ef1b76ebb161354034537ac05c89a34172f7b6e9c8be9f6c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc480471cb427bf10e4f248a24454a98102aaef6bf200026684072b2f5f2581.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418eaf16785cbdd0e5b8fb5d886e2768f4347b8dda669cd924ad0cd857e7e08a.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 14 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 45075184f52be336a8f24e8232251aec1a5890d9a6f0d35cbef2602dca1dafaa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction with no extern calls. SHA-256 over the sorted per-operation ID/hash lists is 0b5d1f4be62738f2ec3f7c2d58a78bf4a984c225f07190229a422d2befcb71ce for source IR, fbb3c6200d5aab32ef48e87fe38ef428e0d44a8d1c99190ad4291a0a3a4f9797 for LTOIR, and 0f8b2b96f9dc634cb2e8c4ffdebff62f8bcebfe00ae08ba7ab2f975ad5ee382f for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; SHA-256 over the sorted per-operation ID/cubin-hash list is ea355e5985a293b10266691d71302aad755879854308203a8f9a3f3bd0ce1303. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `cp_async_ca_16`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 16.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b7c596a91052b75958df78b963338ca733b9b53643eb4b669015b106f926a4.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 99c87c9201c30820920f63624477dfab9046044be0c1e4e9e878a8bd5edd425f.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d4109cb4c2deaedb98fce7f124afb56882fc7ca5ef239a53d86a293d152b465. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact sixteen-byte cache-all copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_ca_4`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 4.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 62367bcc29688f386037306dee01e279908da96b4fbc872a0731f6c1efe3b979.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fc90fb194ad06cc37e4ec41495d29b2ced1dadd23700f4a6742ba5cb274f236c.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61893d39abe94a66af4946061499f48731945913886e7c08b8e5c3761e2e2089. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact four-byte cache-all copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_ca_8`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.ca.shared.global with copy size 8.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b0555ffe6d212958fd651503471d80d0a62030dc35f1dbdac5aedd702be82dd9.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4115fad6c387b8a02ba11bff217ad8d8f244b96f080aad856ea45b4475ee27f8.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2b95d90d47b38d58ceb6c07c46652eac5c8dec1e1ae2ede4f13972c17c1b268. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact eight-byte cache-all copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_ca_zfill_16`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted both register and immediate source-size forms.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4eb30d0f094641ac8dc5af009c2818cf66555f890dc7e6ed4c8e92987f6ee493.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91e8999a4e6f9db5fd5ba08aa79bd58870a77e05e12b06523f182ec6cfb7a14d.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d5bc6dde1a35b6d38bf05e0d3aea8d44520505a79dedfb9bc4fa3f96543e9c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact sixteen-byte cache-all source-size copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_ca_zfill_4`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted both register and immediate source-size forms.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7c8298238217b06520a14baa7dd9017ef6824f9a237ff1c1059c16bcf4a79797.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3022aaec048152a29d81d9b43ad7121669183c87def7ffd6bcc83d52d7040f91.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9928045f3f05dc39ca94c224c2afe7c21092310cdfa83a18da46628fa549c1cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact four-byte cache-all source-size copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_ca_zfill_8`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted both register and immediate source-size forms.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0b171b000534f40633027c1b9b7ab26b739a7ebb686ec8617019f3834cad070.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a01e92a24c61a5a33a96cee208154315907b4538b97345be0e979839c5358c02.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa2dd28dc4c740dac8fce5c55f0002a8bf7e9ab2210bacfd152777eb0e1928b0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact eight-byte cache-all source-size copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_cg_16`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.cg.shared.global with copy size 16.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f32b02e50561865fcbd41af2ab4e6d8c12485b436b14d191cbf5bbf0d248ea82.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f72afd215027bcf67444625672135ba3fbbfadbee3e28f8e90ced82b0321419e.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e091384fbc0457e593996514ad9eeafc04773c58054a8bc85a8c96980c139e5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact sixteen-byte cache-global copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_cg_zfill_16`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the typed AS3/AS1/i32 declaration and canonicalized its argument-memory and pointer attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted both register and immediate source-size forms.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ce9e52734dd2438c5fc127548069d9cb7130610dc9e0752f6598268db9a3a8fe.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5081c35375c4d19aaa528f8abbeb82c57237e9596018318466d212559d1dd65b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a0e1e9bc0608c3d6581f78c2e17d4ef3e71edcd3570f4bde0d9e8f545c8abd6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact sixteen-byte cache-global source-size copy in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_commit_group`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the zero-operand declaration and canonicalized it as nounwind.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.commit_group.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ea4574621afbb04281bfa8ab26db49b9a256ea5a991c63ff2907c9a725aef5a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b2f1831e64e9e04bbe61a19f62e926eebb9e8c7a2e6e7f3fc073d97363c81f2a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 533d9b5b9a9c927d1fe0b23c76106771cdac35407624486b921bcf4c7096811a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact commit-group instruction in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_mbarrier_arrive`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the generic-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3d6f3f642d9e80245cb2f6c76442ccd0d824326e288898d121d5645368edcc8e.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8b4dfc81ee704ec2a8f8e8502b5f5673cafe35607eb28120b57fff0c3b27baf0.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 86913ed12af6b0bd59e448e3b5ccc6823e2f0b6a140146ead578815ea6f7c5ec. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — libNVVM accepted the four-record diagnostic module and produced 2,108-byte LTOIR with SHA-256 70aee00d5928460cc730c2dcb304d14ada723f9ebd8b863894e15d9a556cb5fb.
     - device link on `sm_80, ptx93`: `failed` — nvJitLink rejected the retained kernel because the typed intrinsic had an incorrect argument type and aborted the broken function.
     - backend codegen on `sm_80, ptx93`: `succeeded` — The generated cp_async_small example produced 10,664-byte LTOIR with SHA-256 71f5f67dd2b32ab2e2dd4b61dc640ae0667de68a171bdde713996c187c2e8ba0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 12,320-byte cubin with SHA-256 1fe37bd0a57fe25d8354df11995fd6005e598fb7814b0d30b60cc88ead290d2e; SASS contains two pending-count updates and four ARRIVES.LDGSTSBAR.64 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_mbarrier_arrive_noinc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the generic-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.noinc.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11cd942627957bdbe07521415b591132bac6543a6d37a10286821ca4be498129.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d65ebc7c63cdfa57f0f5aefa206fd07f6004879beb44350640071d5eb97e6625.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ef8193aee88f5efa25293d77c1399be0b83dd294401561177e91f144714072a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — libNVVM accepted the four-record diagnostic module and produced 2,108-byte LTOIR with SHA-256 70aee00d5928460cc730c2dcb304d14ada723f9ebd8b863894e15d9a556cb5fb.
     - device link on `sm_80, ptx93`: `failed` — nvJitLink rejected the retained kernel because the typed intrinsic had an incorrect argument type and aborted the broken function.
     - backend codegen on `sm_80, ptx93`: `succeeded` — The generated cp_async_small example produced 10,664-byte LTOIR with SHA-256 71f5f67dd2b32ab2e2dd4b61dc640ae0667de68a171bdde713996c187c2e8ba0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 12,320-byte cubin with SHA-256 1fe37bd0a57fe25d8354df11995fd6005e598fb7814b0d30b60cc88ead290d2e; SASS contains two pending-count updates and four ARRIVES.LDGSTSBAR.64 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_mbarrier_arrive_noinc_shared`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.noinc.shared.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b62ad0b051d6ccaa435c3686d1a94547c3635490996a003e78abb772b160cc4.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9affe7627501776e21eb8059bdb4bee726a956435720cc6de05ee4d93ef51be1.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 366217f18f9473cf5ccc1d87632a590d8fcf7a8f5508a5bf876c49df1dc939e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — libNVVM accepted the four-record diagnostic module and produced 2,108-byte LTOIR with SHA-256 70aee00d5928460cc730c2dcb304d14ada723f9ebd8b863894e15d9a556cb5fb.
     - device link on `sm_80, ptx93`: `failed` — nvJitLink rejected the retained kernel because the typed intrinsic had an incorrect argument type and aborted the broken function.
     - backend codegen on `sm_80, ptx93`: `succeeded` — The generated cp_async_small example produced 10,664-byte LTOIR with SHA-256 71f5f67dd2b32ab2e2dd4b61dc640ae0667de68a171bdde713996c187c2e8ba0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 12,320-byte cubin with SHA-256 1fe37bd0a57fe25d8354df11995fd6005e598fb7814b0d30b60cc88ead290d2e; SASS contains two pending-count updates and four ARRIVES.LDGSTSBAR.64 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_mbarrier_arrive_shared`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.mbarrier.arrive.shared.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cd2056552d6e2b796c195f31ad768ca175eac447825d819f5957b9fcec07059c.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d59be5c8527b799671b85f003a16a51bbd07448cc2130863d25e7318f2841384.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 23bdc05d4be1c27246a496b4f2ebede69efa8b84d326395adc4d804f16f9eef5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — libNVVM accepted the four-record diagnostic module and produced 2,108-byte LTOIR with SHA-256 70aee00d5928460cc730c2dcb304d14ada723f9ebd8b863894e15d9a556cb5fb.
     - device link on `sm_80, ptx93`: `failed` — nvJitLink rejected the retained kernel because the typed intrinsic had an incorrect argument type and aborted the broken function.
     - backend codegen on `sm_80, ptx93`: `succeeded` — The generated cp_async_small example produced 10,664-byte LTOIR with SHA-256 71f5f67dd2b32ab2e2dd4b61dc640ae0667de68a171bdde713996c187c2e8ba0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 12,320-byte cubin with SHA-256 1fe37bd0a57fe25d8354df11995fd6005e598fb7814b0d30b60cc88ead290d2e; SASS contains two pending-count updates and four ARRIVES.LDGSTSBAR.64 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_wait_all`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the zero-operand declaration and canonicalized it as nounwind.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.wait_all.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1b3dc9f7a7dd7ebd28ad446ba1b8a97a30ffa6065ab3b9bdbb2bcb4377287859.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2dfc842c70441fe57d5a892d34441cf37fedab3955f511c5ef23118632e06924.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 96ab3b4af8cae0f8d415560129805aec75b9e3e8c26b916cda16a42cfc11d581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact wait-all instruction in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cp_async_wait_group`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — LLVM recognized the declaration and canonicalized its operand as immarg and the function as nounwind.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact cp.async.wait_group with an integer literal.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c4b4d445968d5e8db78f64d6903e5dab71d7e0a7ed2edd66bff7dd5443900376.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0604968fa8260d2a46de521f458b9ea4da639127e0671340e531eb5af1e5bea6.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 77ca31b604849a0fbb4a6eeff8a57d146053d576721be6dc2212ec7347f0f1c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact constant wait-group instruction in the combined classic cp.async production example and produced 8,052-byte LTOIR (SHA-256 e048104e8dac3507a23165041436116c2d0a696f0939df71663c5c6dbced1199).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 7,712-byte sm_80 cubin (SHA-256 91e7704506ae29972952a49c14621ef1246e4144619066b489c622fb6782d5ae); cuobjdump showed all eight LDGSTS copy forms and the LDGDEPBAR/DEPBAR control sequence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_f16x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rn, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 23c48425228bb21090b442cbcb63b58874efaf6cc05dde2ad7549aff23e3f8ce.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7c94821ab335a6f02f2b3ca52d1e0c2428d1d26aac5e7c62852ef35cbfcba8e4.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 009a800094f29371a5f120862eb39d45b097aaa5715ce2e7ac26a43ec3665773.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fa4b5787be34f9a5c70cf30bc30034cd847065603dd1be2ed9ac938a537224d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact cvt.rn.f16x2.f32 at compute_80 as part of the combined six-form inline-PTX probe and produced 2,164-byte LTOIR (SHA-256 a0617e3e49b1df2beb5f575ae8750e0b62a57ef403db4b793dc77b5e1ef715f7). This proves only the production inline-PTX route; it does not claim typed libNVVM support.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 95af39e283c3ad359158c7ed7dcca98a43ebfac1a6a7c49f7c10081224218726). cuobjdump shows the same six F2FP forms as the LLVM/ptxas path. A compute_75 LTOIR reached nvJitLink but failed terminal PTX compilation because every form requires sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_f32x2_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rn, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7b66943f3e72c942117544102ede74b7d46537dc9c8f09cb4c800b1a8150b25.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 260159c4d2bde9577b33387b20e93e5e1e00605c7ce69ce157c5c5046bbaabf0.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 625e5da377d73334b52ec4bdd6bc6702e15b196ca54350f0b070cb61e09b60ff.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 160baca629eded02d3ab3a52f8e77fc3220065d5054f9025a7e6519830c17c0c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact cvt.rn.bf16x2.f32 at compute_80 as part of the combined six-form inline-PTX probe and produced 2,164-byte LTOIR (SHA-256 a0617e3e49b1df2beb5f575ae8750e0b62a57ef403db4b793dc77b5e1ef715f7). This proves only the production inline-PTX route; it does not claim typed libNVVM support.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 95af39e283c3ad359158c7ed7dcca98a43ebfac1a6a7c49f7c10081224218726). cuobjdump shows the same six F2FP forms as the LLVM/ptxas path. A compute_75 LTOIR reached nvJitLink but failed terminal PTX compilation because every form requires sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_f16x2_e4m3x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b295a2577d8a238ef3098f7cb0e4fc309f9e5f1177f9f02ae766101f9d70cc67.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdceb080a053c25340804b87375d5fc8693a5fc8eaa461399d438982c1cd179a.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5682cacdfec0f3e7fe02db10b1665b3b88ab176b4db26738c965155f31ba0dbf.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e06c823f79624ff5a141d8abb9d054774f71faca396e9b031841bc0bf0b42c9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_f16x2_e5m2x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a572f8ac9426c4acabd244748fbcabfffc19a7afbc3eb6b74b087a7397d0ca5.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 182d55254c86ae42a4eaaf4db2c9829f771eb0f23f6abaa92f773fd9910b607b.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7d96da22340431d9ae279f1ea4d4573bca35a171e7fc6199b5893b540a93424.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 52e1205cad104ec1b2f8206e28faaa263dc716532dc29e75bf488e1761a57d68. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_relu_bf16x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.relu.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.relu.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rn.relu, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e02262263ec043530dec4469b22cfd6e831bc597d97b59e20132167985ae8e75.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c797710407ad5b5c6425fc3ae32587456573e2442fb1398a784e61ab025c7f6b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 336ac20640c8b6451d754f6afed9479d89a22cd5c47b6606e57f251337b35e0c.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d75e6605feeb1d636c46023a4abb8c4a61c3cf7ba8cd9a35ef81ef224d5cf806. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact cvt.rn.relu.bf16x2.f32 at compute_80 as part of the combined six-form inline-PTX probe and produced 2,164-byte LTOIR (SHA-256 a0617e3e49b1df2beb5f575ae8750e0b62a57ef403db4b793dc77b5e1ef715f7). This proves only the production inline-PTX route; it does not claim typed libNVVM support.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 95af39e283c3ad359158c7ed7dcca98a43ebfac1a6a7c49f7c10081224218726). cuobjdump shows the same six F2FP forms as the LLVM/ptxas path. A compute_75 LTOIR reached nvJitLink but failed terminal PTX compilation because every form requires sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_relu_f16x2_e4m3x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e12dd4763bbc1533bd40687021b8319bd843af8e62fc1fefe8212335310449ab.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e0264fce6d34633b3225bd3f48723a6efaa11b18a738168e34c460dfe2b16754.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a2341dac9ffd42fede2aefd66f576cac3e464c0b6b81c38773128832579eafa.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e9f96f0cf12b836acbe8449eb09c0b680298cd1898fed30eaf8c5ede4caab7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_relu_f16x2_e5m2x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b32d14ca011f505a526f0fecf85660bd93e1bbc147e728896df311e7f2b1176b.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f69dc7503144fa82436f92c37e77254ebf11ca61ffdcf492ce8b6ace6ae00234.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dff12609bfb2cbd5258c5d20e2e1c8b781090475d157dfe239cdbfc5edd9c204.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41e9e6b8a2f160a5e3ca517eeaac8b418cf9e29219e4e3e84987424aac1d5f33. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_relu_f16x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted exact cvt.rn.relu.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact cvt.rn.relu.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rn.relu, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f4b2d11b6f47357e52b5da2bd8374d6043945e2d3c7d28f0d9bce741658cbbfb.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a886266d7776914c367d748d72bd9b1032db99c5eb5366f902a3b5b8e63f2add.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 644949d91a98e2c1b36b125ca7211d181b486d46e6b0335c823f0f78de9860cc.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 55c5bcea1fc302ac84979c69abc416ce1bf8a413b5886c1cdd0627e8d59039b8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact cvt.rn.relu.f16x2.f32 at compute_80 as part of the combined six-form inline-PTX probe and produced 2,164-byte LTOIR (SHA-256 a0617e3e49b1df2beb5f575ae8750e0b62a57ef403db4b793dc77b5e1ef715f7). This proves only the production inline-PTX route; it does not claim typed libNVVM support.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 95af39e283c3ad359158c7ed7dcca98a43ebfac1a6a7c49f7c10081224218726). cuobjdump shows the same six F2FP forms as the LLVM/ptxas path. A compute_75 LTOIR reached nvJitLink but failed terminal PTX compilation because every form requires sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_relu_satfinite_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 8.6 / sm_100+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,181-byte opaque-pointer NVVM IR fixture (SHA-256 4a6b6f02fd2af26b07cce84c876b37378eb153dcbf109f1683f95d5a79d544dc) to 991-byte PTX 9.3 (SHA-256 5b12b177c8a6e39c4c2010ca9f6d65f1e54dc69b622eb38c50c9a54e245e2f47) and 2,748-byte LTOIR (SHA-256 33aff65e7fe18658f37c2d3e77cf0909b6b67890b1471fa0e65b27dcff71abe1), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the 510-byte exact sm_100/PTX 8.6 fixture (SHA-256 5201f213e5e3a049b0adce8b8d7a18ef55f99aaee11210feb111f4ccc96f2bac) and produced a 5,536-byte cubin (SHA-256 30efab89a6a6f44df3d434ec54b318edb08a5c9b078ad4b107454ba8f7deb582). It also assembled the libNVVM-emitted PTX to a 5,536-byte cubin (SHA-256 4cfc63ecdd67d02b26ab53bf62fa6f17c5bbe9fb3361dc0f3a5c420d9b536acd). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 5,536-byte sm_100 cubin (SHA-256 5dbb5e71fcad9c659eb627dc8fd57f980455d3164ec63c978574102cb84cff9a). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_relu_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 7.8 / sm_90+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,203-byte NVVM IR fixture (SHA-256 4b5b61cbde895ad4cca3777fb86c33c06e96d35b70d9fc589d3e0126201dd222) to 956-byte PTX 9.3 (SHA-256 aca3fe16c949e7614f988c423f06912fd05a431babafe2def5af07b03abaaea9) and 2,040-byte LTOIR (SHA-256 d0affd25b909632cee088d988a7778bde56d065edbd5c1324c3e1a8cf957bb23), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 469-byte exact sm_90/PTX 7.8 fixture (SHA-256 7455770182c5f46203de2170777b7bce100e445a41562bbc34945256325fa93a) and produced a 3,704-byte cubin (SHA-256 33f8941a912c4580054123f3ce55e599143c13cdfa161f0b3de1c55158a1780f). It also assembled the libNVVM-emitted PTX to a 3,704-byte cubin (SHA-256 1302e6a89ef1629ad728d8459ca785a8dc0a8901231869bbb56c0435e78922bb). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 3,704-byte sm_90 cubin (SHA-256 7758fa36cb80e9d0f3bd35c3084f710b4dd47870ffb3e9c4f4f1f378a2beefab). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_e4m3x2_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 777bd7a5362a935cb9b45c99eb5c7912d8296fd322f232be5feb4dcb71afe958.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8843a1e5b9fee11892601ac9688d35a5195bdd8c88122bb6ab5b4fa21cf76e29.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d344820320e9db759a6db9eebd370929159ce71d483a76cb70f846d27abf426.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c5390039d4cfeed4d5708005fc89938d3c0b2b9ff580d4d4d86a004deb5bb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_e4m3x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb627e6f8e2597ba015a8fc53f58f4956dee58`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,184-byte NVVM IR fixture (SHA-256 1f48f717966e2249dc5f2d82492c9fd945cc7bf1a87becef38df48881995fd88) to 1,068-byte PTX (SHA-256 371b97f7ec8b94fbe36b4406dc227fe62e4df8dbba8541b512b5e7c998e13323) and 2,080-byte LTOIR (SHA-256 f29c9bdfd76cb78bdd90783f31c97d9a984379f171df011ed97926faaebc5596). All four forms retained high, low PTX source order.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all four forms at sm_89 and produced a 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced the 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). The same route at sm_87 failed terminal compilation with the sm_89 requirement for every form. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_e5m2x2_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b6fcb8ea52dc533221fc3a953835ab9fb2cdb8a3c82a1dc9527dc2b980cb096.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6707ccb77a07e8b004032798bbfd07afba3af30d15568ac2c3d08d84d8b10b3.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6064f89491e6f84f9ac56c6373eec145a9b8d55507b30a4ee2ffcbe2deed7dd1.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f3263d6f3b707aebc45c8e4608b974d23fd5769c7d9886716e35456201ee254a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_e5m2x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb627e6f8e2597ba015a8fc53f58f4956dee58`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,184-byte NVVM IR fixture (SHA-256 1f48f717966e2249dc5f2d82492c9fd945cc7bf1a87becef38df48881995fd88) to 1,068-byte PTX (SHA-256 371b97f7ec8b94fbe36b4406dc227fe62e4df8dbba8541b512b5e7c998e13323) and 2,080-byte LTOIR (SHA-256 f29c9bdfd76cb78bdd90783f31c97d9a984379f171df011ed97926faaebc5596). All four forms retained high, low PTX source order.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all four forms at sm_89 and produced a 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced the 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). The same route at sm_87 failed terminal compilation with the sm_89 requirement for every form. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_relu_e4m3x2_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a17f8309a9b1436ceae14d34278cbc25993c1607850ec87962d7fef9fd6c3215.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11284b3464e72006408b832095a07a2ee8683336b367d675109a20b5289db239.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 82c1a28789b05b8df6d92cc52932afdc99dc761cb2bcc77c7d85e2b3a1da1e24.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 086de3bfab9d5f0aa20f2c59eaab134fad12e7e9b1023f8deffaf01a48966bd7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_relu_e4m3x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb627e6f8e2597ba015a8fc53f58f4956dee58`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,184-byte NVVM IR fixture (SHA-256 1f48f717966e2249dc5f2d82492c9fd945cc7bf1a87becef38df48881995fd88) to 1,068-byte PTX (SHA-256 371b97f7ec8b94fbe36b4406dc227fe62e4df8dbba8541b512b5e7c998e13323) and 2,080-byte LTOIR (SHA-256 f29c9bdfd76cb78bdd90783f31c97d9a984379f171df011ed97926faaebc5596). All four forms retained high, low PTX source order.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all four forms at sm_89 and produced a 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced the 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). The same route at sm_87 failed terminal compilation with the sm_89 requirement for every form. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_relu_e5m2x2_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(<2 x half>) and four <2 x half>(i16) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 2,785-byte IR (SHA-256 0f0b893b3fdc9d13bbd03894017d82295121adbf15942d5fae5c0d2d7a9f1452).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,815-byte fixture (SHA-256 76eab1d5b8e485d46480392bfbcd8414b8c84ddd9f82cbf642da68a84452ce57) to 4,830-byte PTX (SHA-256 323876d5eb482939160e6718240f5358876c7abf50de4f000f64cee27f64c449) containing all eight exact conversions. PTX 8.0 and sm_87 were rejected with 'Cannot select: intrinsic %llvm.nvvm.f16x2.to.e4m3x2.rn'.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95) retaining the E4M3/E5M2 base and ReLU forms in both directions. sm_87 was rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,366-byte inline fixture (SHA-256 51a75a316c1334d60ef82ea936bab8ac02dfe57562890c494c017110471f0fa7) to 5,102-byte PTX (SHA-256 a45759f6aa9ead00769d3ec786eb46407d3152295ce0fb57743fa0244c2a7bd6). Every form used the reviewed $0, $1 order with =h,r constraints when packing and =r,h constraints when unpacking.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte cubin (SHA-256 7d86e0e31bcf5c2c446f9426a20db5359dfb53f0bd1bffcc9032aa890f9e4c95), byte-identical to the typed-NVVM cubin, confirming both routes emit the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2a8ce890aff0fa2df126aba76cf0dcc3fbda114de759ad215043e6ad0d400d23.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab076ae9d2de65d0ab3caad272a118bd426726141914f2b539a30a616d655e4e.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cf21d0dd82dfc3a278f16fc60cf6772892bc2faee051ae9dd92b23f6e934f314.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c8fd72479f8c912c05e01db30ae4db47a935ecec7dbd2ac3b9c80689a4639f2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,774-byte NVVM IR fixture (SHA-256 177071f5289fef1bade95e29f12afa689c91dc8f4c6f4d89cca810527c1317da) to 1,628-byte PTX (SHA-256 54595aa338e74747cf5ec117a0570f112c673a3d0266b7d2a5d850d3ba6c8ba4) and 2,268-byte LTOIR (SHA-256 dae644d035c13e6afec18e39d26ba16c8326caecab2d6425158b563d3d91a4ba). All eight forms retained their single packed source operand.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all eight forms at sm_89 and produced a 3,624-byte cubin (SHA-256 c761f544931d83935d10c7965037bad1536fd8d1f30cea4ae52f3826b4e8821e). The same PTX emitted for compute_87 was rejected with "Feature 'cvt.e4m3x2.f16x2' requires .target sm_89 or higher". Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 1,464-byte cubin (SHA-256 aadc245b2a8c0d63d2a3d5337bff8cb7c778022514b8f32d1e1efd61bf91ac3f) from the sm_89 LTOIR. Linking that same LTOIR for sm_87 failed with 'LTOIR arch newer than target arch'. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_relu_e5m2x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Pinned LLVM canonicalized the four i16(float, float) declarations to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in 1,724-byte IR (SHA-256 d6a1f164920582f756e24a01e4ac1ccd87042b45fb7202bcabfc27dad23e095a).
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,128-byte fixture (SHA-256 c34fcffc96185a5a3d746fb99435955858780171b842b88b1939e94469ca0fe7) to 977-byte PTX (SHA-256 8da714495549c455c6a77d168e5d193463037164645fa90f8cc505dccbc1adb1) containing all four exact conversions. PTX 8.0 and sm_87 were rejected.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 8758c825ff05674628381bbf19cfe4ba99f5077971ce6066e863c292fb42b82a) retaining E4M3/E5M2 base and ReLU forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_89, ptx81`: `succeeded` — Pinned LLVM lowered the 1,022-byte inline fixture (SHA-256 5919ef9b476cd61b2e276cd2726e6db9cf443ac6dced739366284450a50087fa) to 1,305-byte PTX (SHA-256 ac6134577dc1f9083d0373e5883bd028f2b7d2cd27891c600598725eb10c1701). Every form used the reviewed $0, $2, $1 order and =h,f,f constraints.
-    - PTX assembly on `sm_89, ptx81`: `succeeded` — CUDA 13.3 ptxas produced a 3,368-byte cubin (SHA-256 f238c85c61f29667b999a4bc0fadcbfafa85b980558687e1b6ba6e1f3217482d). PTX 8.0 and sm_87 failed with the expected floor diagnostics. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd31e276f07818fdd0a11d4334bf456235366a9c46422edc617ef0140d6997ab.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa496a4aa988e4fd9ab8baf8b1ad392ebe27bbacb8c7baafc198d96c6756993a.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9050d525b077779e4e5e0ddfa8ab75608b22adab5731337e1925cb992440558c.
+    - PTX assembly on `sm_89, ptx81`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6dded5e03b1ce7eb2e2bfe0e88f4f3a77fb6b5c87f3e0fa2f5312cb1358b0b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.1 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb627e6f8e2597ba015a8fc53f58f4956dee58`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,184-byte NVVM IR fixture (SHA-256 1f48f717966e2249dc5f2d82492c9fd945cc7bf1a87becef38df48881995fd88) to 1,068-byte PTX (SHA-256 371b97f7ec8b94fbe36b4406dc227fe62e4df8dbba8541b512b5e7c998e13323) and 2,080-byte LTOIR (SHA-256 f29c9bdfd76cb78bdd90783f31c97d9a984379f171df011ed97926faaebc5596). All four forms retained high, low PTX source order.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all four forms at sm_89 and produced a 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced the 3,368-byte cubin (SHA-256 e52e9372da7213161b9c6987cb718ca9c508e325c91a6f76538eff35fda911e7). The same route at sm_87 failed terminal compilation with the sm_89 requirement for every form. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_satfinite_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 8.6 / sm_100+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,181-byte opaque-pointer NVVM IR fixture (SHA-256 4a6b6f02fd2af26b07cce84c876b37378eb153dcbf109f1683f95d5a79d544dc) to 991-byte PTX 9.3 (SHA-256 5b12b177c8a6e39c4c2010ca9f6d65f1e54dc69b622eb38c50c9a54e245e2f47) and 2,748-byte LTOIR (SHA-256 33aff65e7fe18658f37c2d3e77cf0909b6b67890b1471fa0e65b27dcff71abe1), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the 510-byte exact sm_100/PTX 8.6 fixture (SHA-256 5201f213e5e3a049b0adce8b8d7a18ef55f99aaee11210feb111f4ccc96f2bac) and produced a 5,536-byte cubin (SHA-256 30efab89a6a6f44df3d434ec54b318edb08a5c9b078ad4b107454ba8f7deb582). It also assembled the libNVVM-emitted PTX to a 5,536-byte cubin (SHA-256 4cfc63ecdd67d02b26ab53bf62fa6f17c5bbe9fb3361dc0f3a5c420d9b536acd). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 5,536-byte sm_100 cubin (SHA-256 5dbb5e71fcad9c659eb627dc8fd57f980455d3164ec63c978574102cb84cff9a). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rn_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 7.8 / sm_90+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,203-byte NVVM IR fixture (SHA-256 4b5b61cbde895ad4cca3777fb86c33c06e96d35b70d9fc589d3e0126201dd222) to 956-byte PTX 9.3 (SHA-256 aca3fe16c949e7614f988c423f06912fd05a431babafe2def5af07b03abaaea9) and 2,040-byte LTOIR (SHA-256 d0affd25b909632cee088d988a7778bde56d065edbd5c1324c3e1a8cf957bb23), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 469-byte exact sm_90/PTX 7.8 fixture (SHA-256 7455770182c5f46203de2170777b7bce100e445a41562bbc34945256325fa93a) and produced a 3,704-byte cubin (SHA-256 33f8941a912c4580054123f3ce55e599143c13cdfa161f0b3de1c55158a1780f). It also assembled the libNVVM-emitted PTX to a 3,704-byte cubin (SHA-256 1302e6a89ef1629ad728d8459ca785a8dc0a8901231869bbb56c0435e78922bb). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 3,704-byte sm_90 cubin (SHA-256 7758fa36cb80e9d0f3bd35c3084f710b4dd47870ffb3e9c4f4f1f378a2beefab). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rna_satfinite_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx81`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the fresh 628-byte probe to 634-byte IR (SHA-256 51e877e213ff97519d0a1c4080afee97c3e4b208b88ab2071b214a36428c77f4) with the imported f32-to-i32 signature and canonical intrinsic attributes.
-    - backend codegen on `sm_80, ptx81`: `succeeded` — Pinned llc lowered the fresh probe to 634-byte PTX (SHA-256 be00dc0e407a9f71d14c5b73a5a4c53b06e4430ad20f888e5595499b25c348ad) containing cvt.rna.satfinite.tf32.f32.
-    - PTX assembly on `sm_80, ptx81`: `succeeded` — CUDA 13.3 ptxas accepted the fresh sm_80/PTX 8.1 output and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a38ab8cd30065c8554fc11f7b6f155740eb533d290525b57a45a4b339465234.
+    - backend codegen on `sm_80, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2819c3052033ee04b85149f6891347ea860e7bacc511a0e7edb1d5fb4f36cd2.
+    - PTX assembly on `sm_80, ptx81`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e8c2a92a43cdb367ea7856ca832f6fea44f45613b70f89f2acd2b45c4d8b35b4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 8.1 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 712-byte NVVM IR fixture (SHA-256 53576c4b24a86d9113541fef9ac285909ebf80d3760faafc6a716b0e2bbb723f) to 635-byte PTX 9.3 (SHA-256 106b44a5be000e6c9e34f1b2314c6b1ff0d7ebd48309e960edb675f47c63714e) and 1,824-byte LTOIR (SHA-256 13e6d38c7eb00722eb78918bdcfb4042ea14c8cfb03fca08adbd79556637eb75), retaining the exact instruction and =r,f constraints.
     - PTX assembly on `sm_80, ptx81`: `succeeded` — CUDA 13.3 ptxas accepted the 292-byte exact sm_80/PTX 8.1 fixture (SHA-256 c4cbf838c25be5aa3f314b02af5e0319480ce1e8ac14ae6b3c2dd1de57169d4b) and produced a 3,240-byte cubin (SHA-256 4a947924e65d45b1fd736ce16701e20fdc415da6976ee4be24a4ae5ef6ee6a96). It also assembled the libNVVM-emitted PTX to a 3,240-byte cubin (SHA-256 52f9ca2dc160df01842137e25185ad0ca49f79da4754843c64356ab543fdad52). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 3,240-byte sm_80 cubin (SHA-256 a1462e268ef5c361faaeaf2dafca16dfb4eb1bee56fcce16589792cdb3cf350d). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rna_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the fresh 598-byte probe to 584-byte IR (SHA-256 2448daf9235fa4db9287c592309f7496856e9cf31929014a569c70813978d9c8) with the imported f32-to-i32 signature and canonical intrinsic attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc lowered the fresh probe to 565-byte PTX (SHA-256 19695568a4fde5c5847f25e5103f5f4aecf6b96dd9f408028fee66222fe8c82f) containing cvt.rna.tf32.f32.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the fresh sm_80/PTX 7.0 output and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7fc68746b0ee106bcb3270f8a02cb83cd0a4e85363c0854f82546717a33bf72d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5355197dbd45ff46716deac34ba4d8de1084ee4374f537af9a67dd2fb47b6c5.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8bf2670d9bea4e04e196ffe05f2954a5d9f8beefc57a231af094444b20348c3a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 702-byte NVVM IR fixture (SHA-256 10498f999cc34865430b1f8d64021483334eb3c83d32fb2bb08a1c28b284b7c7) to 625-byte PTX 9.3 (SHA-256 d43ba7bc4ca91e3bfeac5ee8b4e3ceffe56943bae28a17e2f5440ff30500397e) and 1,808-byte LTOIR (SHA-256 de830d41c3ff7ae593112eed2361814d717cc66332aad9146a9336911d6e14cf), retaining the exact instruction and =r,f constraints.
     - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 282-byte exact sm_80/PTX 7.0 fixture (SHA-256 fa8e77156d6f3b4fa592ac6652d220f33ffe1df3e23d80715c8715ba7ef80f1b) and produced a 3,240-byte cubin (SHA-256 dd4762da0d14241138620d6e9e08e09deb39fc20f6eb124d6bcdf9c7295cf4fe). It also assembled the libNVVM-emitted PTX to a 3,240-byte cubin (SHA-256 9af0070a67901f2348ab176bf45e4570664d84bc0312ab6d797ebf2c70326f15). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 3,240-byte sm_80 cubin (SHA-256 0d4de7970f5141c056194d046e10dd4b455e84bccd02fe82d8dcea808c42e9e4). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rz_bf16x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted exact cvt.rz.bf16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact cvt.rz.bf16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2bf16x2.rz, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb6bcfe8642c6981459470ff9b577231bb7c5575846df5a6b524e3ed834bfc4c.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9dcafda6629d6de7301da44bf1e7690a5d1e8131d595cc975ea4e8b68a61de5d.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd97b8bfbf98ce29b538e25417e266f1d5c29d1d18b0789af1dd439fe4431d90.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3118ee80f44de9320ac595fac8d8638286f81f4c483f50b7fb363dc5e94c7db6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact cvt.rz.bf16x2.f32 at compute_80 as part of the combined six-form inline-PTX probe and produced 2,164-byte LTOIR (SHA-256 a0617e3e49b1df2beb5f575ae8750e0b62a57ef403db4b793dc77b5e1ef715f7). This proves only the production inline-PTX route; it does not claim typed libNVVM support.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 95af39e283c3ad359158c7ed7dcca98a43ebfac1a6a7c49f7c10081224218726). cuobjdump shows the same six F2FP forms as the LLVM/ptxas path. A compute_75 LTOIR reached nvJitLink but failed terminal PTX compilation because every form requires sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rz_f16x2_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80`: `succeeded` — The selected Rust LLVM 22.1.2 backend canonicalized this declaration to nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) in the combined 2,683-byte canonical IR artifact (SHA-256 172ae44a3b3f55425bb296f9f485ea08fe9f51dada6721d5a5fbf8b61cfd6d60).
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted exact cvt.rz.f16x2.f32 once in the combined 1,286-byte typed PTX artifact (SHA-256 263e8020dcdc7ccbb25baeefba06104abde3cac5100e98d40f9304746992ef97). Passing high then low preserves the public first-argument-low lane convention.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the combined 1,286-byte typed PTX module at sm_80 and produced a 3,624-byte cubin (SHA-256 f12457bfc7cc44d96014d876f4705a6ddee14db8471828df74dec91d060ff929). cuobjdump contains all six distinct F2FP packed-conversion forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact cvt.rz.f16x2.f32 in the combined six-form production probe, with the public low/high arguments printed as PTX high/low. An independent typed call to llvm.nvvm.ff2f16x2.rz, invoked with the kernel high value followed by the low value, emitted the same exact instruction and PTX operand order even though the normalized imported record has no named selections.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,462-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 c92d92589eadec753288e74985a1775e3f822765bffb6a88746c4a31ccaf7b6d). cuobjdump contains all six distinct F2FP packed-conversion forms. PTX 6.5 rejected every form as requiring PTX 7.0; sm_75 rejected every form as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2bccf7085ce3ccc76e323145e0cd1e86fda391d0411adc51c1eea4ce806a2f03.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5790119795c6ad9cbfc37ff239e9a123b44c05bde5d38fc5675565933d23a6ff.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6833e6724886aa52db6de2c812bc6e0e2a7216e33ba07501701559eb4c68cf5c.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 342a17c23f404e77ec2d3a0e7b1b008dd8b6face8fdbd672e4dda87c91b17e7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact cvt.rz.f16x2.f32 at compute_80 as part of the combined six-form inline-PTX probe and produced 2,164-byte LTOIR (SHA-256 a0617e3e49b1df2beb5f575ae8750e0b62a57ef403db4b793dc77b5e1ef715f7). This proves only the production inline-PTX route; it does not claim typed libNVVM support.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 95af39e283c3ad359158c7ed7dcca98a43ebfac1a6a7c49f7c10081224218726). cuobjdump shows the same six F2FP forms as the LLVM/ptxas path. A compute_75 LTOIR reached nvJitLink but failed terminal PTX compilation because every form requires sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rz_relu_satfinite_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 8.6 / sm_100+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,181-byte opaque-pointer NVVM IR fixture (SHA-256 4a6b6f02fd2af26b07cce84c876b37378eb153dcbf109f1683f95d5a79d544dc) to 991-byte PTX 9.3 (SHA-256 5b12b177c8a6e39c4c2010ca9f6d65f1e54dc69b622eb38c50c9a54e245e2f47) and 2,748-byte LTOIR (SHA-256 33aff65e7fe18658f37c2d3e77cf0909b6b67890b1471fa0e65b27dcff71abe1), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the 510-byte exact sm_100/PTX 8.6 fixture (SHA-256 5201f213e5e3a049b0adce8b8d7a18ef55f99aaee11210feb111f4ccc96f2bac) and produced a 5,536-byte cubin (SHA-256 30efab89a6a6f44df3d434ec54b318edb08a5c9b078ad4b107454ba8f7deb582). It also assembled the libNVVM-emitted PTX to a 5,536-byte cubin (SHA-256 4cfc63ecdd67d02b26ab53bf62fa6f17c5bbe9fb3361dc0f3a5c420d9b536acd). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 5,536-byte sm_100 cubin (SHA-256 5dbb5e71fcad9c659eb627dc8fd57f980455d3164ec63c978574102cb84cff9a). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rz_relu_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 7.8 / sm_90+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,203-byte NVVM IR fixture (SHA-256 4b5b61cbde895ad4cca3777fb86c33c06e96d35b70d9fc589d3e0126201dd222) to 956-byte PTX 9.3 (SHA-256 aca3fe16c949e7614f988c423f06912fd05a431babafe2def5af07b03abaaea9) and 2,040-byte LTOIR (SHA-256 d0affd25b909632cee088d988a7778bde56d065edbd5c1324c3e1a8cf957bb23), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 469-byte exact sm_90/PTX 7.8 fixture (SHA-256 7455770182c5f46203de2170777b7bce100e445a41562bbc34945256325fa93a) and produced a 3,704-byte cubin (SHA-256 33f8941a912c4580054123f3ce55e599143c13cdfa161f0b3de1c55158a1780f). It also assembled the libNVVM-emitted PTX to a 3,704-byte cubin (SHA-256 1302e6a89ef1629ad728d8459ca785a8dc0a8901231869bbb56c0435e78922bb). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 3,704-byte sm_90 cubin (SHA-256 7758fa36cb80e9d0f3bd35c3084f710b4dd47870ffb3e9c4f4f1f378a2beefab). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rz_satfinite_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite canonical IR SHA-256 values are af7f0adb4b4af6b1780d0d197f00bf32ee51e9d8dbd81d6b97e5f3ec101df15b, 4d5f18c4b4f68957c8f79beb9328183d80dace302ad00ad85d45dc0576006462, 67c65855b54636a577a1cb45317a39b4254ddec8799e777b0256533c8d4cab7f, and 244675d8ce634cbca9a0882bb019bd8f0b09ba07634b3c91b619c0c184d029b2.
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn.satfinite, rn.relu.satfinite, rz.satfinite, and rz.relu.satfinite PTX SHA-256 values are e81778bdf814fec0843aae928af55096f48b421e6097b4399d520dfa26c38a68, 65c3aa4699184bdf18d5ae60ba23729c1157f7d76336ec921487fe54d49dc486, 39c5f1828614f6c8ae09d5b70e893344d749991eae512e009fdfeb492d8cedcd, and 277d21990b992ac107687f16fbe77cd31fed1252d311832becd819b39e03f672.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_100/PTX 8.6 outputs. Each produced a 2,240-byte cubin with SHA-256 590812e6b719b50435534538a5329aa69b6bd2fb0c8684250ce9c4c6484828ce. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a223dcc57720b04066b9682e737de671731dd27b040da9cf1ad56f0d4580b71.
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cbc701ff1e0404eb48e20f2e43a6ecbb77778fc6a0c04054e4b1ea67eb5a8bdb.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ed180fa33eddd3aaa2d55a8906d5a35288c5d46c10a3c53e6acad31e814aa7c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 8.6 / sm_100+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,181-byte opaque-pointer NVVM IR fixture (SHA-256 4a6b6f02fd2af26b07cce84c876b37378eb153dcbf109f1683f95d5a79d544dc) to 991-byte PTX 9.3 (SHA-256 5b12b177c8a6e39c4c2010ca9f6d65f1e54dc69b622eb38c50c9a54e245e2f47) and 2,748-byte LTOIR (SHA-256 33aff65e7fe18658f37c2d3e77cf0909b6b67890b1471fa0e65b27dcff71abe1), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the 510-byte exact sm_100/PTX 8.6 fixture (SHA-256 5201f213e5e3a049b0adce8b8d7a18ef55f99aaee11210feb111f4ccc96f2bac) and produced a 5,536-byte cubin (SHA-256 30efab89a6a6f44df3d434ec54b318edb08a5c9b078ad4b107454ba8f7deb582). It also assembled the libNVVM-emitted PTX to a 5,536-byte cubin (SHA-256 4cfc63ecdd67d02b26ab53bf62fa6f17c5bbe9fb3361dc0f3a5c420d9b536acd). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 5,536-byte sm_100 cubin (SHA-256 5dbb5e71fcad9c659eb627dc8fd57f980455d3164ec63c978574102cb84cff9a). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `cvt_rz_tf32_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all four fresh probes. The canonical IR SHA-256 values are 448216daaa2897bf4093e7da556a07b30d6d4a01b2a7707bc8900d280a57d849, 6ea4dc8b3afde70862652b0cd6c2489181da5f52ac40472dd4d59ce1db71bf8d, fa6ca3587009d60b14a17e125bff4baaab9619340d8aba26aafc13ede76b3e0b, and c3ac997a03440a0526eadf63a961c6470d1eca7361f1977f2c9ee78ff52a5af6.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — Pinned llc emitted the four exact instructions in fresh PTX. The rn, rn.relu, rz, and rz.relu PTX SHA-256 values are 73bfe99ccf3356a128d9e7a0de6fb38c40e8d2b1c79c36048f2d88fd28519c66, 1069fa0f1148a8bb978130aafb882b5f33d8896e2898a6ac161f8cd8911a821e, aa249634cf5ffefe5fdb978df3317d1675de529485e859839087936ffa8a558e, and 556259c1eeb2d2a9020a4987a920e00ff8ae29bf63b6e44eddf8459b3ac90431.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted all four fresh sm_90/PTX 7.8 outputs. Each produced a 1,720-byte cubin with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f6dc640b0524a3e1389398b36eff75ef8dae55003578c682753f637372771203.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b333c2a24bba1df7f85b531f3e9dcf1f20488aaa64dcb0aa68cb7ec587bbe88a.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 36b7cb98609bf1bcad4274ea91bdecdf1ea55281e80f200738f80e2608c71937. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-conversion` at PTX 7.8 / sm_90+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the fresh 1,203-byte NVVM IR fixture (SHA-256 4b5b61cbde895ad4cca3777fb86c33c06e96d35b70d9fc589d3e0126201dd222) to 956-byte PTX 9.3 (SHA-256 aca3fe16c949e7614f988c423f06912fd05a431babafe2def5af07b03abaaea9) and 2,040-byte LTOIR (SHA-256 d0affd25b909632cee088d988a7778bde56d065edbd5c1324c3e1a8cf957bb23), retaining all four exact instructions and =r,f constraints.
     - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 469-byte exact sm_90/PTX 7.8 fixture (SHA-256 7455770182c5f46203de2170777b7bce100e445a41562bbc34945256325fa93a) and produced a 3,704-byte cubin (SHA-256 33f8941a912c4580054123f3ce55e599143c13cdfa161f0b3de1c55158a1780f). It also assembled the libNVVM-emitted PTX to a 3,704-byte cubin (SHA-256 1302e6a89ef1629ad728d8459ca785a8dc0a8901231869bbb56c0435e78922bb). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink 13.3 consumed the fresh LTOIR and produced a 3,704-byte sm_90 cubin (SHA-256 7758fa36cb80e9d0f3bd35c3084f710b4dd47870ffb3e9c4f4f1f378a2beefab). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rm_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rm_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rm_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rn_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rn_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rn_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rp_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rp_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rp_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `div_rz_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `dp2a_s32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_61`: `succeeded` — LLVM recognized the declaration, immediate selector, and no-memory attributes.
-    - backend codegen on `sm_61, ptx50`: `succeeded` — The selected Rust LLVM backend mapped false to exact dp2a.lo.s32.s32 PTX at the native floor.
-    - PTX assembly on `sm_75, ptx50`: `succeeded` — CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_61`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cf14ce2498510d2ccdf8e26062d9dab33a2f018ab44d7402f731ee721edd9586.
+    - backend codegen on `sm_61, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5004901d9a7fa9fc719218c051ffd4362bcef5a21c7bcca397e74fdb199f3733.
+    - PTX assembly on `sm_75, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c5f55a84bfd467a639f8fe3020d3fa05a6616160125d05ecd44cf631c78a954c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 5.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx50`: `succeeded` — CUDA 13.3 libNVVM accepted the exact low-half inline-PTX route at its supported compute_75 floor.
     - device link on `sm_75, ptx50`: `succeeded` — libNVVM produced 1,984-byte LTOIR; nvJitLink produced a 3,240-byte cubin; cuobjdump showed IDP.2A.LO.S16.S8. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `dp2a_u32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_61`: `succeeded` — LLVM recognized the declaration, immediate selector, and no-memory attributes.
-    - backend codegen on `sm_61, ptx50`: `succeeded` — The selected Rust LLVM backend mapped false to exact dp2a.lo.u32.u32 PTX at the native floor.
-    - PTX assembly on `sm_75, ptx50`: `succeeded` — CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_61`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6bba537191d91e4c7f577889e7832f879421a5a8b021a98982af2c068ecd5ac5.
+    - backend codegen on `sm_61, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ad295ebd24c5fa94a1990f0e676049b7232bf0421f6419e2230f054510b87ede.
+    - PTX assembly on `sm_75, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aef5395fd4ebfc37a0a4fb468bb18c83a363528cd307bfeb7ed1218c5fd84fc2. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 5.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx50`: `succeeded` — CUDA 13.3 libNVVM accepted the exact low-half inline-PTX route at its supported compute_75 floor.
     - device link on `sm_75, ptx50`: `succeeded` — libNVVM produced 1,984-byte LTOIR; nvJitLink produced a 3,240-byte cubin; cuobjdump showed IDP.2A.LO.U16.U8. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `dp4a_s32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_61`: `succeeded` — LLVM recognized the declaration and added its no-memory and speculatable attributes.
-    - backend codegen on `sm_61, ptx50`: `succeeded` — The selected Rust LLVM backend emitted exact dp4a.s32.s32 PTX at the native floor.
-    - PTX assembly on `sm_75, ptx50`: `succeeded` — CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_61`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 de5ba8b2a4efbc2ca94148badabcdf0d50913a97a1663fb623592433f484e88e.
+    - backend codegen on `sm_61, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9c42148d5c5a275ca9ece93315e98eb0e9508a0b4e8cf0eba8cc761268f29c42.
+    - PTX assembly on `sm_75, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d9b7b74a68441f7f40d480ab430166c946704ea11caf1a7e9951b4726130bc65. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 5.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx50`: `succeeded` — CUDA 13.3 libNVVM accepted the exact inline-PTX route at its supported compute_75 floor.
     - device link on `sm_75, ptx50`: `succeeded` — libNVVM produced 1,984-byte LTOIR; nvJitLink produced a 3,240-byte cubin; cuobjdump showed IDP.4A.S8.S8. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `dp4a_u32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_61`: `succeeded` — LLVM recognized the declaration and added its no-memory and speculatable attributes.
-    - backend codegen on `sm_61, ptx50`: `succeeded` — The selected Rust LLVM backend emitted exact dp4a.u32.u32 PTX at the native floor.
-    - PTX assembly on `sm_75, ptx50`: `succeeded` — CUDA 13.3 ptxas accepted the native-floor output when assembled for its supported sm_75 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 5.0 / sm_61+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_61`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4565681a3a6f6f64ea89b63e96c44c1aef2d594b01f3c695eaa4d35147c250f.
+    - backend codegen on `sm_61, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eec368e7d9686a4072001c8362bc69ea91a9b25fb950b50232244c96e72c0887.
+    - PTX assembly on `sm_75, ptx50`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 159beb8496be094d32d2c6421e8993b91ee8e0d8b787fa5313af73ea5af66eac. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_61 +ptx50, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 5.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx50`: `succeeded` — CUDA 13.3 libNVVM accepted the exact inline-PTX route at its supported compute_75 floor.
     - device link on `sm_75, ptx50`: `succeeded` — libNVVM produced 1,984-byte LTOIR; nvJitLink produced a 3,240-byte cubin; cuobjdump showed IDP.4A.U8.U8. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `dsmem_read_u32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-cluster-memory` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM emitted both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 690-byte source had SHA-256 ad8a94dfa560d523bfe8cf963f8779132c960e1e6e5717e080e72819936f446a; the 1,007-byte PTX had SHA-256 005a7f90c8a3bd97a99792ce8a181d43c28fb318cd20dce6068fcb349800a951.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 3,712-byte cubin with SHA-256 1780aee43cd691bd7e3a79f4bc4b2d163e5cdfc11e08441ae318dfb091a5057a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-cluster-memory` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418a46a6a0279c0e2c7391fcccae3b3bec1a660a3ffad28b1df2a7f63df297d6.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bf4f5d63cf41eae708cc3087c76cbea8b9812ba26f1c915646ef4b0205c51760. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-memory` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 1,222-byte source had SHA-256 b966d8131b036df01f435a9dab89559917d5ac23d6e26d0adfc23a959d899cdc; the 2,208-byte LTOIR had SHA-256 a8006b43c0aacce7c48c04a1d3599127db090a21b4f2ed7cddc1d676a6243728; the 922-byte linked PTX had SHA-256 e8d9eea75eec5d6179fab2ba58d900c14bc5ed3af1c9ffe2c39f6e7c9fc798d0.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,840-byte cubin with SHA-256 5fc93334fcfa3fe77cee54dd6470bdecb562c8ecd205425b64df9d81a95ae091. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `dynamic_smem_size`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 4.1 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx41`: `succeeded` — Pinned llc emitted both size reads in 655-byte PTX with SHA-256 4ab2630ee33912457ed45c1cd516ae0797e062fcba7123db9bac352c67280b3a.
-    - PTX assembly on `sm_75, ptx41`: `succeeded` — CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 ee3358b59d2b596f1b276d983c3149d58c384872ee5edad482cf20a99183e7b1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 4.1 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx41`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24cc1707db68cd33a711956ac50c8c65ab3e84eb5cf71a00fc15160d770b645d.
+    - PTX assembly on `sm_75, ptx41`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7feb7ce1442470cb7ca3f613338c9fa901e16b23bc621a84ac3aa46a8583088d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx41, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.1 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained globaltimer, smid, gridid, warpid, dynamic_smem_size, and total_smem_size in 2,028-byte LTOIR (SHA-256 e00f26c9c773f5ea24198cda4a679a0de54ceafb3a13be5190b4dcf0b475f864) and emitted 1,135-byte PTX (SHA-256 4cb7846fcc9a10b3d26ccbab61c1936f26ae6255911d637cf6875fbf4e6d3bea).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 60456548b55c17dbef4e6a269c3eacb20298af44406ed81eec5657362501cdda; cuobjdump retained globaltimer, virtual SM, grid, warp, and shared-memory size reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `elect_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-elect` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as accepted the 1,080-byte source with SHA-256 e83b2cd7839b679e3d23da509c817d79108c84ce4eabefc70282dd2065cf69a1 and produced 2,100-byte bitcode with SHA-256 77043e86454445b633d31ec577a6c27c5a2675b960421a0739a621de3896a9e9.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — LLVM 22 emitted exactly elect.sync %r2|%p1, %r1 and elect.sync %r3|%p2, -1 with no extern calls; the 826-byte PTX had SHA-256 86e7fcebf74be1752a172688c87ec3bf042a296f7846df731b0385ad9145bf23.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas produced a 4,328-byte cubin with SHA-256 f13f6d2c6aa0971a28ecf01f5008727bc513b0acef9cdd075973f03f30c15876. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-elect` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d6be8db0355a531bcc20af99d6b2ab9a8015c92ce61965522dec78614fd03d1c.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 de0e95f3672e5c96d15e4fd737d9225940692889c495211037f7b4f99d3cddb4.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 192c7ae696bf472409f022926d3c74f98784499ab59d769d1565e6605976c7d0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-elect` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `failed` — Direct CUDA 13.3 libNVVM PTX compilation rejected llvm.nvvm.elect.sync with 'Callsite was not defined with variable arguments' and NVVM_ERROR_COMPILATION. The 525-byte source had SHA-256 50c9d2f3d137bc6a626c3f26618fe2e27f4fbf5611793390f98d49ab8b365e68.
     - device link on `sm_90, ptx93`: `failed` — libNVVM produced 1,760-byte LTOIR with SHA-256 8ece269aaf4f669c1b16eb297b6fa7729de34d449cc17b9ca48149db87c28f7e, but CUDA 13.3 nvJitLink rejected it with the same broken-callsite error and produced no cubin. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained both exact inline elect.sync blocks with no extern calls. The 1,915-byte source had SHA-256 e59f441c92eaa1b3194fbfaa77a7f9614161dec2464feb7d3e9cf36f1c27b622; direct 515-byte PTX had SHA-256 eceb98cddfddfdf09a72710cd13462af745eb44eb8b15b5082e560a58a5845d5; 2,120-byte LTOIR had SHA-256 7a8dd8f0a26a975352370eb00246f79f9c111d59d8d9847efd785c3264e4a92b.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,560-byte cubin with SHA-256 fe782c618044a7c8e6777ad12d29b31e8e3c9ebdbb6c4eabf6319f3045b67397. Direct ptxas also accepted the retained PTX and produced a 3,560-byte cubin with SHA-256 dca3adbbd0f75cc7d15f3b10e94f621d47fb9d658d3a3ae623d1b1ee769be76e. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `envreg1`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 2.1 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75`: `succeeded` — CUDA 13.3 libNVVM verified all six intrinsic declarations before compilation.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,316-byte LTOIR (SHA-256 57b3175e40f2a217ede37a7b8dc3aadce172260ec9ed5459eca6b2d7023711bb) and exact 879-byte PTX (SHA-256 dfac22a9564622fa9b2e9ef64f97a1a50e0e23c853f9734e2aa7c5b8a22ad8eb).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 276fc6a71465f9db688bafa30cd6f85084f291f9fab7a7155e090aa3db1173d3; cuobjdump retained clock and device-constant register reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `envreg2`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 2.1 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75`: `succeeded` — CUDA 13.3 libNVVM verified all six intrinsic declarations before compilation.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,316-byte LTOIR (SHA-256 57b3175e40f2a217ede37a7b8dc3aadce172260ec9ed5459eca6b2d7023711bb) and exact 879-byte PTX (SHA-256 dfac22a9564622fa9b2e9ef64f97a1a50e0e23c853f9734e2aa7c5b8a22ad8eb).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 276fc6a71465f9db688bafa30cd6f85084f291f9fab7a7155e090aa3db1173d3; cuobjdump retained clock and device-constant register reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ex2_approx_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx70`: `succeeded` — Pinned llc selected ex2.approx.f16 for sm_75 with +ptx70; PTX SHA-256 21a1d2fe684434a6c71b2454ff0a26463002def6ab7c09e00252a70d9770b7df. The same probe fails selection at sm_70 or +ptx65. llvm.nvvm.ex2.approx.ftz.f16 and llvm.nvvm.ex2.approx.bf16 fail selection; llvm.nvvm.ex2.approx.ftz.bf16 selects at sm_90 with +ptx78.
-    - PTX assembly on `sm_75, ptx70`: `succeeded` — CUDA 13.3 ptxas assembled the 640-byte sm_75/PTX 7.0 output into a 1,464-byte cubin with SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cee684a507daecda40b46b40a8de9a2f2582dfcf5aae88b3d844b49e614c44e7.
+    - PTX assembly on `sm_75, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0fd052b8dd9f1821ac995eba3175320422af7c36dcdf0fc99190a1b2819de6e6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.0-libnvvm-12.0.140-ex2-f16` at PTX 7.0 / sm_75+: status `validated` (`CUDA compilation tools, release 12.0, V12.0.140; based on NVVM 7.0.1`, SHA-256 `91aa5019c144d0049bd482b9fd897f73a39f4ac56dbc1e8b4d59b259cfc50860`)
     - backend codegen on `compute_75, ptx80`: `succeeded` — CUDA 12.0 libNVVM verified and compiled the 6,524-byte production NVVM IR module for compute_75. With -gen-lto it emitted 3,880-byte LTOIR; its PTX mode emitted 4,523-byte PTX 8.0 retaining ex2.approx.f16 exactly once. The originating build observed IR/LTOIR/PTX SHA-256 values 08c7c2f7a6bf6c0c4dc8dbccaf088061befb053ec2987e9663844195b874a649, 1894a0a1fe26304d2e36794350800dd9aad909a5c0d068ef996355a0dbf30ccb, and d04240dc1acd4bcb4d6b2a8ae7ecdee9de30092be444dd64b7fd3e02641f70f9. Those intermediate hashes are build-context dependent because Rust crate-disambiguator text appears in helper symbols; independent rebuilds reproduce the sizes, target, PTX ISA, instruction, and the exact final cubin, whose hash is the reproducible artifact identity.
     - PTX assembly on `sm_75, ptx80`: `succeeded` — CUDA 12.0 ptxas assembled the 4,523-byte PTX into a 4,264-byte sm_75 cubin with SHA-256 4fd44330c1bae0181ae46648955f541615143f43bfe067243a53d48e7b872bfe. Tool `/usr/bin/ptxas` reports `Cuda compilation tools, release 12.0, V12.0.140; build cuda_12.0.r12.0/compiler.32267302_0` (SHA-256 `332c1b822fcab8d86db02236973deac3661cd4b122e0e54158ae38fd7f6d1def`). Artifact `cubin`.
     - device link on `sm_75, ptx80`: `succeeded` — cuda-oxide's cubin materializer used CUDA 12.0 nvJitLink's versioned 12.0 API to link the 3,880-byte LTOIR into a 4,264-byte sm_75 cubin with SHA-256 4fd44330c1bae0181ae46648955f541615143f43bfe067243a53d48e7b872bfe. cuobjdump disassembly contains MUFU.EX2.F16. Tool `/usr/lib/x86_64-linux-gnu/libnvJitLink.so.12.0.140` reports `CUDA 12.0 nvJitLink V12.0.140` (SHA-256 `b6c94451e08cf91fb6d9ffc5c10cc7de6f936996733ad56e220a607bb08580f9`). Artifact `cubin`.
 - `ex2_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified and compiled a dedicated typed-pointer NVVM IR fixture holding the 3 appended variants (ex2.approx.f32, ex2.approx.ftz.f32, tanh.approx.f32) for compute_80. The emitted PTX 9.3 output retains each instruction exactly once with no extern calls; PTX SHA-256 9d2a3b28989cf40d76816090ce6051c2347f46d8bbbd6fd7fb5b479977e5d7cc.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the appended-fixture PTX into an sm_80 cubin with SHA-256 134a765e19dd2267cf41934ad4ffa3ec0a6f54c43406d4da3102adfb9334e955. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the appended fixture with -gen-lto to a 1,848-byte LTOIR (SHA-256 6c6fcf6eb5ddef3cf3ea605c0a866c8cc4db63bae1df7abca521ff507283b56d) and nvJitLink 13.3 linked it into a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632) with all three appended instructions retained. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ex2_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified and compiled a dedicated typed-pointer NVVM IR fixture holding the 3 appended variants (ex2.approx.f32, ex2.approx.ftz.f32, tanh.approx.f32) for compute_80. The emitted PTX 9.3 output retains each instruction exactly once with no extern calls; PTX SHA-256 9d2a3b28989cf40d76816090ce6051c2347f46d8bbbd6fd7fb5b479977e5d7cc.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the appended-fixture PTX into an sm_80 cubin with SHA-256 134a765e19dd2267cf41934ad4ffa3ec0a6f54c43406d4da3102adfb9334e955. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the appended fixture with -gen-lto to a 1,848-byte LTOIR (SHA-256 6c6fcf6eb5ddef3cf3ea605c0a866c8cc4db63bae1df7abca521ff507283b56d) and nvJitLink 13.3 linked it into a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632) with all three appended instructions retained. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fence_mbarrier_init_release_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fence_proxy_async_generic_acquire_shared_cluster_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx86`: `succeeded` — The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor.
-    - PTX assembly on `sm_90, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97.
+    - PTX assembly on `sm_90, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.6 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fence_proxy_async_generic_release_shared_cta_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx86`: `succeeded` — The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor.
-    - PTX assembly on `sm_90, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97.
+    - PTX assembly on `sm_90, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.6 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fence_proxy_async_shared_cta`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_acquire_cluster`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_acquire_cta`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_acquire_gpu`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_acquire_system`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_release_cluster`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_release_cta`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_release_gpu`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fence_proxy_tensormap_generic_release_system`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 91aa9fe5869bcabda7787109170938f2be211ebafe71fab7b6c66efcb6b90d2d for source IR, 2f67163d6bde1d1fbf75ff36dae6ca9077f8af5b5870f30537cd553ba9515e5d for bitcode, and 8fc72ca7a768c0dc82369f9a6f06ffca5ca2255c169936b1e8d0eb406f463944 for canonical IR.
-    - backend codegen on `sm_90, ptx83`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 db8e952f291c518f4b61d347ed39363e010647e369d38f1404ffd3dbae969e79.
-    - PTX assembly on `sm_90, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 f927dcc01687588f4ffee8c0aa6f70756cfdb95c5c51a7d314b815f9051ed62f. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a196edad03b8970b9fdb568238486ad53a6b32434e4ea689ae2c566573d36c7b.
+    - backend codegen on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95e027558053f12dc197731ea6153a81059c1db1ec8889ae2e31efc9a2d9d41b.
+    - PTX assembly on `sm_90, ptx83`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34517e49af777f5face8b082c3a6e0adbda8ca3fa1cd46fcfc6eb970ebe44727. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is a5c34eb855405098b2c46729f2b4ed6e9c0e4dbb348900a2fb7161b06e34af1d for source IR, 5112b46e3a6e47e6ba4bef5b5330973f1c0621b04b63d3f0d670db0c55da724a for LTOIR, and 81662318dd9dace3a3522010e8555402e42dbcc1b48ec3c7d8650d28a11cdbe7 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 e8ae2d8892057e1f7895023d878daf75101d77896c304557d87decafa135e23a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `fma_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact fma.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f847702ab3d3db2051359e9ba8869735a57a5cf5b92a2901b154340dc14cc080.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa10765bb36d761080b94f2c0788a77e387b12032c06504740e5ff756239cda5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact fma.rn.bf16x2 route at compute_80 as part of the combined seven-form probe and produced 2,256-byte LTOIR (SHA-256 a9394a9ba0813c36f65bfc2636d7e80b27ab2ffcb6ded14a420d4127a824028d).
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,624-byte sm_80 cubin (SHA-256 1c8ff2eeb9e58f8b3defc22ed929cd48ec2e2f2beab30747f937a9a7869d74d3); cuobjdump showed HFMA2.BF16_V2, HFMA2.BF16_V2.RELU, HMNMX2.BF16_V2, sign/absolute HFMA2 forms, and F2FP.BF16.PACK_AB. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact fma.rn.f16x2 inline PTX in the combined four-form PTX 6.0/sm_70 probe. An independent typed call to llvm.nvvm.fma.rn.f16x2 emitted the same instruction at PTX 6.0/sm_70 and at its native PTX 4.2/sm_53 floor; sm_52 failed exact intrinsic selection.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e1cf777d5d76a7579b7c840ac1250cd7961888126aeef92b619e2bfcf5718aa.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f962661dab009881f23af0cdf33aed0bdb5ea13629f790897da04f271d8cc305. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_75 as part of the combined four-form probe and produced 1,932-byte LTOIR (SHA-256 edb864f3999af5c06558f9dde87cd506ac0c2188cb6a71b19fbf8b33d6b5e5ae). The instruction floor is PTX 4.2/sm_53; compute_75 is this libNVVM profile's floor, and compute_70 is rejected as an unsupported profile option.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,368-byte sm_75 cubin (SHA-256 5ca3d1ca1d9c56802bb78bdad8f5e1138b17dc4dc9f96a7b1e9572eb4010d50c). cuobjdump SASS contains HFMA2, HADD2 add/sub forms, and HMUL2, matching the LLVM/ptxas path. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `fma_ftz_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80a7eb25c335cf849bc1e0be9ed1e391b9d311a1573d38be927d6aa35c8b4aa3.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83de32412117576acec7391b12d4c1e5f2f026f4fed9281689ec4a87e9b1deaf. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 746-byte fixture (SHA-256 178f2b3dbe2bd55b7f329c53abc033302ceaaa88ac7e4a472ad3ed8d4d8cd503) to 1,486-byte PTX 9.3 (SHA-256 9df510c14e1753de1ff549bef0b31be13a080cf8aa5bddf7701d1a4a57f280cf) retaining all 3 exact instruction head(s).
     - PTX assembly on `sm_75, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_75 cubin (SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM emitted 1,936-byte LTOIR (SHA-256 0cd7e34b43ca73e0bcff828b2f9b6461d122976da6fd775133b092e7690a469d) under -gen-lto. nvJitLink produced a 1,464-byte cubin (SHA-256 aa90a4ef126034727fa94a004f35ec9d37642da982becc0efc4b82ab19499dcb), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_ftz_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `fma_ftz_relu_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 1 exact instruction head(s) in 682-byte PTX (SHA-256 ee283e72de7ddec2258b4eb628ca180889b67d9f79d13ba926719724bec1c01f) lowered from the 196-byte fixture (SHA-256 fa242c0f862cc57936c53949b30d4ce59871e59cde0349570a44122c3a164006). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the module at sm_80 and produced a 1,464-byte cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93c9c1a8aaa9ac2bee924f4b1eee264e0aa38c8c6f01c04bddf3bfd58d169bc4.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 582782aeec6a6286313dd28330e3de2a23eeaa07c479ee6271c831e0c8f9d1e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 443-byte fixture (SHA-256 2945f1e57f54d050923c2112dc596fe3f6700e68dd833525528fc839065db648) to 633-byte PTX 9.3 (SHA-256 03d4146479a9f72b237545bc07910b9a9af2c3de56f7575940f31cdc7fa15c2b) retaining all 1 exact instruction head(s).
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 1,584-byte LTOIR (SHA-256 e4177fb92e02319d1ba7af5c06481ec10888852bdf7bae3b59e362a804d83cf4) under -gen-lto. nvJitLink produced a 1,464-byte cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_ftz_sat_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9bed831d23f91ca28a7fa178103ea679a4765ec587f918986874e52b21d00da1.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6a185d53a7cf04592fc205ee4ef0926f252fa57f5c50f1ee510c7aa6210578ee. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 746-byte fixture (SHA-256 178f2b3dbe2bd55b7f329c53abc033302ceaaa88ac7e4a472ad3ed8d4d8cd503) to 1,486-byte PTX 9.3 (SHA-256 9df510c14e1753de1ff549bef0b31be13a080cf8aa5bddf7701d1a4a57f280cf) retaining all 3 exact instruction head(s).
     - PTX assembly on `sm_75, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_75 cubin (SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM emitted 1,936-byte LTOIR (SHA-256 0cd7e34b43ca73e0bcff828b2f9b6461d122976da6fd775133b092e7690a469d) under -gen-lto. nvJitLink produced a 1,464-byte cubin (SHA-256 aa90a4ef126034727fa94a004f35ec9d37642da982becc0efc4b82ab19499dcb), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_relu_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact fma.rn.relu.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 903e4771d5a308e61a37282dd1f236481315e6519d5c06dbfd19c6c47f97acfd.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 adc9face773753523147665457351aa23a8b43da9e697a565109e15870383010. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact fma.rn.relu.bf16x2 route at compute_80 as part of the combined seven-form probe and produced 2,256-byte LTOIR (SHA-256 a9394a9ba0813c36f65bfc2636d7e80b27ab2ffcb6ded14a420d4127a824028d).
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,624-byte sm_80 cubin (SHA-256 1c8ff2eeb9e58f8b3defc22ed929cd48ec2e2f2beab30747f937a9a7869d74d3); cuobjdump showed HFMA2.BF16_V2, HFMA2.BF16_V2.RELU, HMNMX2.BF16_V2, sign/absolute HFMA2 forms, and F2FP.BF16.PACK_AB. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_relu_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67da91ea322e2d60af10838e8a20890d99c38e07262abe9a6848ea5e1fbbf2fb.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 871b4ac65868041bcc0f028deaaac2793f5df9db4518fdbfaa03580e9c26414a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_80 as part of the combined three-form upper probe and produced 1,860-byte LTOIR (SHA-256 54983753f6c883e98ddef424042a704c0e38f32fcff6620ef12cf4734543d549). The instruction floor is PTX 7.0/sm_80.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 f70d50b154c66b2dc59909f29f449274a86d8300048982b8b1aa17b276193fe0). cuobjdump SASS contains HFMA2.MMA.RELU and both HMNMX2 predicate forms, matching the LLVM/ptxas path. A compute_75 LTOIR negative reached nvJitLink but failed terminal PTX compilation because all three forms require sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rm_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rm_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rm_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rm_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rm_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rn_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rn_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rn_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rn_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rn_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rp_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rp_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rp_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rp_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rp_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rz_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rz_ftz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_rz_sat_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized these declarations in the 44-form arithmetic and 20-form add proof bundles. Their manifest SHA-256 values are f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba and 0ea3fb5767d9527b171de379b7ce5fdaeaee8c433e278cda0f67bb4973e1ac60.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 16 selected instructions across two probes. The PTX files had SHA-256 13bae6359a033e218c3d8cda6a05d34ef017f94ffb4127204c341003a596b2e5 and c0904b539a5d409f456b9a5e6145e4bb47d1e94a565680a892276fdab3603e9b.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted both probes and produced cubins with SHA-256 f15cdd950be3a925782a0958c3522bd2d26c6f75742bb41779a3b8c841d472d7 and 79ea7082bde55f2d1b36c9d53b861e85c2df8683ec01927803a376f547b0adf7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f71dcf802590ec4ed143c3050ac0f77955b009d0b3fb8d7028de32d14d68d954.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5cca2d6d4c97eb2cf997702b416cd5e1cbbd098204583c3f3828ea301d19336.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 16 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 584f9cf756c55613f53213b15ad53e6d8ce2ff767c5d711ed3d360c1a8b01be0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `fma_sat_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — Pinned llc retained all 3 exact instruction head(s) in 1,855-byte PTX (SHA-256 3cec642b59813f21bc9f86d2d39e00dac4defde30827f4bcd0ba7361810ec441) lowered from the 499-byte fixture (SHA-256 be3dbf17dd609b89c4f12fa6e8532498a1914e1c5ddbbd99c37ae4fc75db80f2). Independent typed calls to the exact llvm.nvvm symbols emitted the same instruction heads.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the module at sm_75 and produced a 1,464-byte cubin (SHA-256 0869784cef1bfc816ca2564c97cb80d49645941a74c4658c829baaee24f8f30b). The admitted floor is sm_70, but CUDA 13.3 ptxas no longer accepts that target, so assembly is checked at sm_75 as it is for the other PTX 4.2 packed-f16x2 forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7ccf304fffd63a678dec5f5775eb1fbdc60e0a295b7b2a9e878339cf8a0922a.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ad17848cb5677cf0a68e6e3eb435b41a165f6ef1cbaf69b0fa3e2b4d8bbfdcfe. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 746-byte fixture (SHA-256 178f2b3dbe2bd55b7f329c53abc033302ceaaa88ac7e4a472ad3ed8d4d8cd503) to 1,486-byte PTX 9.3 (SHA-256 9df510c14e1753de1ff549bef0b31be13a080cf8aa5bddf7701d1a4a57f280cf) retaining all 3 exact instruction head(s).
     - PTX assembly on `sm_75, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_75 cubin (SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM emitted 1,936-byte LTOIR (SHA-256 0cd7e34b43ca73e0bcff828b2f9b6461d122976da6fd775133b092e7690a469d) under -gen-lto. nvJitLink produced a 1,464-byte cubin (SHA-256 aa90a4ef126034727fa94a004f35ec9d37642da982becc0efc4b82ab19499dcb), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `globaltimer`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30`: `succeeded` — llvm-as and llvm-dis produced 571-byte canonical IR with SHA-256 05f5c3e1cb6db2e75012139ce279738ff495d4073aacfd429b50bdfde6153f44.
-    - backend codegen on `sm_30, ptx32`: `succeeded` — Pinned llc emitted the read in 548-byte PTX with SHA-256 a78d9cae78923fa3a3be7909f47e29aedb6bfa663681173ec7a9aed44e8f4bfe.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_30 PTX and produced a 2,856-byte cubin with SHA-256 70b45763f48528285ad1525d565230adce99b4abc193a0f6302ca82be86024b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 397d8179a2161540732e23d2358db3f5b9d44664430b730aac91c4dc83e5cded.
+    - backend codegen on `sm_30, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7610b182b5f89de1c960aeb1cac1a4aff14a652935f535dc6e90da6fb8a211e8.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5001111b2c815febac9426aed7a4b9a51ab752be9b2d97985f60a745e49cd0a7. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 3.1 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained globaltimer, smid, gridid, warpid, dynamic_smem_size, and total_smem_size in 2,028-byte LTOIR (SHA-256 e00f26c9c773f5ea24198cda4a679a0de54ceafb3a13be5190b4dcf0b475f864) and emitted 1,135-byte PTX (SHA-256 4cb7846fcc9a10b3d26ccbab61c1936f26ae6255911d637cf6875fbf4e6d3bea).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 60456548b55c17dbef4e6a269c3eacb20298af44406ed81eec5657362501cdda; cuobjdump retained globaltimer, virtual SM, grid, warp, and shared-memory size reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `grid_dependency_launch_dependents`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 73f4f353452e427acdb94ba15dbab4b8a6147137bf7e33decdbf8e41d03152e2, bitcode SHA-256 e666c4c454382bd351d83c3c8ca347bdd7d3bb9e24bb8e6643f306a4893f6ec1, and canonical IR SHA-256 7d1e8a9808ea508ca317b88b06284570bad89943a66e2dec4f7e8c7608688f94.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — LLVM emitted the exact instruction in PTX SHA-256 97504b6bc748ddc18d1e82bded39ecfefb288bf088e5afeca6c519f3a4f900cb.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b0abd382879e45371de563a83219bb7d4d07b5f9f42e1be22a6f4d677203678.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e1f0e1a7641d1458f0fa9d84525e830a2dbb84f58fab281009a922c356df6eec.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 306ce9690dd2d976940425ec7052d1f0cdb3b0cee03e8507345d328dde684796. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 7.8 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 55e66456c6d94738f891265e35df2dc4c562f70204de5974900471c8d90141b0 to LTOIR SHA-256 a756d76960b6510a63454230169f254a36918faa5fffc4defd94eccabc937854; nvJitLink emitted matching PTX SHA-256 9a50bac1217347e320d682ac35f580cf815ac14dbd4e11753325b4ca9d1f93d7.
     - device link on `sm_90, ptx93`: `succeeded` — nvJitLink produced cubin SHA-256 b58003d0006957a49ab1f4c5e305fdb686b4271f1644f1ef58114cd41670a8bd; cuobjdump decoded the launch operation as PREEXIT. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `grid_dependency_wait`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 d7f02e7d83c4693e1821d37b444a9dabb20710f703938b6bf6043b5f5bf48e90, bitcode SHA-256 f1cf4c3dca46407bea36ac7957f7da7909fa6405c8f6da0fc914ab6f5277ab9a, and canonical IR SHA-256 d2d4106872992bf9ca9d7b8b147a7a7ebc6a515b5ee748b9c1cb022eacd92450.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — LLVM emitted the exact instruction in PTX SHA-256 9b54ea0566c66befa6ff9e703c91eba251fa5c0fb9eb477a07c391fd9556045c.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 36680cc0d56219dd30aae173b2d38de4e5141c96daf9543af43a89fc3616e84c. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e63a89f7a0e5ef9325815f030775cc2f52a995704e695e7f6f6bc0a9d29e7e2.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 351547f724188c103868fd7cf963ba564b216c62db91f2a2a24a8685feb892a6.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 10bee9dcaa765674caa7cfcc9059aa263e5cf7b95fcb8fbdca3bee1f03e2b589. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 7.8 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 f8fdc53d19dcba3c9334cc2fe1e3048fb24b2e30d8d09a45d75fcff19c29167d to LTOIR SHA-256 326bc0c859d9b9014f23e13f1da999091e1cc6872e15ec827fa81092d0e6e4d5; nvJitLink emitted matching PTX SHA-256 a90a113865f2ba8c53c1f2b0fee84ef277e3e70bb0cea5653d0fe9ee90ca6246.
     - device link on `sm_90, ptx93`: `succeeded` — nvJitLink produced cubin SHA-256 25b03a5e3721c020006dc2ad615fe0a4a5f8034548a3f3e4bc5454f577fee18c; cuobjdump decoded the wait operation as ACQBULK. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `gridid`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_30, ptx32`: `succeeded` — Pinned llc emitted the full-width gridid read in 582-byte PTX with SHA-256 ee2f3fba8b51cd03bace8fccd9e3a76817240b7eb6e69e29943df3838b730159.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 39c56630ff2afdadeb521c2a56023f289a2a39f0d39b0d01e2a5b3d9171dedd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_30, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fb0578258494bd20b8dc20c6fe2243954b382980ac39ab5577a752f9af9a20e5.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0236c2ff2b8a9c74dc5ef689d53ae080fbc93f5d23cee0b18105663b14080f9c. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 3.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained globaltimer, smid, gridid, warpid, dynamic_smem_size, and total_smem_size in 2,028-byte LTOIR (SHA-256 e00f26c9c773f5ea24198cda4a679a0de54ceafb3a13be5190b4dcf0b475f864) and emitted 1,135-byte PTX (SHA-256 4cb7846fcc9a10b3d26ccbab61c1936f26ae6255911d637cf6875fbf4e6d3bea).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 60456548b55c17dbef4e6a269c3eacb20298af44406ed81eec5657362501cdda; cuobjdump retained globaltimer, virtual SM, grid, warp, and shared-memory size reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m16n16_x1_trans_b8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -8944,28 +8944,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -8998,28 +8998,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9052,28 +9052,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m16n16_x2_trans_b8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9106,28 +9106,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9160,28 +9160,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9214,28 +9214,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n16_x1_b8x16_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9268,28 +9268,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n16_x1_b8x16_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9322,28 +9322,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n16_x2_b8x16_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9376,28 +9376,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n16_x2_b8x16_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9430,28 +9430,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n16_x4_b8x16_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9484,28 +9484,28 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n16_x4_b8x16_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 recognized every exact .p3 declaration and its imported convergent readonly contract in /tmp/blackwell-ldmatrix-probe.ll (4,153 bytes; SHA-256 1fac276e40ad836eb70cc7b8001adbf31fb29e1ef3791c95917cf878b2b8d211). Literal _p64 and _p32 remain part of the LLVM symbol before the .p3 suffix. sm_101a at PTX 8.6 and sm_101f at PTX 8.8 emit successfully, but both abort instruction selection at PTX 9.0; the minimum-only catalog therefore excludes both non-monotonic spellings.
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100a -mattr=+ptx86 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100a-ptx86.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 4516ae48c9beb579d9ab14a3c781c93ae6603a3795db1f10505fb1dde9b7a3de.
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_100f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_100f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 79db57b73a47960e0c10dba3d016b6f473f8243a63e67f5e4aa2a70df9ba7087.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 10011ae80ba6639ff6ca7a85bab67dc1bdd4e5f0e6eb39dde6dc0cc02e967877.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_103f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_103f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 a8ab25dbf7f266f8a3334ac464b6ccb8340616f69f08b9387bbc54e3cd42f0bc.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110a -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110a-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 c88212d07d98aaf04a4ee3f7c33ead0a5485b49e2af4f923540206cf4816a7c8.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_110f -mattr=+ptx90 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_110f-ptx90.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 38cb04adff09a634b51c8910694a68ffac1f761fd52ddbef5a57dc14990bf961.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120a -mattr=+ptx87 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120a-ptx87.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 48506c1e4897a379b2d678548307f08fb36ca3758d4780d363ea79d6072042e5.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_120f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_120f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 853aeeaecd418af27e5ac678b672e504e305ba52bdc01163f867b33f57440c85.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121a -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121a-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 dcf1f4f71b3f6287f0075c40cda014140dd5702f02434337c6b0048b6cfa7087.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Command: llc -O0 -march=nvptx64 -mcpu=sm_121f -mattr=+ptx88 /tmp/blackwell-ldmatrix-probe.ll -o /tmp/audit-bw-min-sm_121f-ptx88.ptx. Output: 7,658 bytes, 12 exact heads, SHA-256 ed9c9cbd5ab7fdc5c14af451b408888760c83bc6c210a809e30a57f4e1f98555.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100a /tmp/audit-bw-min-sm_100a-ptx86.ptx -o /tmp/audit-bw-llvm-sm_100a.cubin. Output: 2,248 bytes, SHA-256 563593a85aec2eb7e9d33e111c641d3cf046e2896aa3db426bb3b398b126a4c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_100f /tmp/audit-bw-min-sm_100f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_100f.cubin. Output: 2,240 bytes, SHA-256 b4f706e96475dcf14db9f87a167d90028485c65cc03d7214638f805824da6a3f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103a /tmp/audit-bw-min-sm_103a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103a.cubin. Output: 2,240 bytes, SHA-256 acef9b2aa67c1217e718e2139ab90080308ee4559e1e7cfb4f76ca29dc00c543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_103f /tmp/audit-bw-min-sm_103f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_103f.cubin. Output: 2,240 bytes, SHA-256 ec3d701370495952cd96ba35b6d6f32d763301cfa99226495c3bcaf40ffdd560. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110a /tmp/audit-bw-min-sm_110a-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110a.cubin. Output: 2,504 bytes, SHA-256 f5a2c7e3f46f5ecf2ac1c938a1a6bc90b52604126d1addb041a2508307a00303. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_110f /tmp/audit-bw-min-sm_110f-ptx90.ptx -o /tmp/audit-bw-llvm-sm_110f.cubin. Output: 2,504 bytes, SHA-256 1f8b803976dde7dd514095f6c40e6e881df50987592a9f88b2dbefac3f8598c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120a /tmp/audit-bw-min-sm_120a-ptx87.ptx -o /tmp/audit-bw-llvm-sm_120a.cubin. Output: 2,240 bytes, SHA-256 9da5f12601952bb54b8b757e61495eb86e824bb51088b0785afa27310760c7d1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_120f /tmp/audit-bw-min-sm_120f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_120f.cubin. Output: 2,240 bytes, SHA-256 9be687b81f3179f4ce921fef3507d4d4e3d8593ca668f2299e62d1985e85d255. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121a /tmp/audit-bw-min-sm_121a-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121a.cubin. Output: 2,240 bytes, SHA-256 c388839ac7be175dafe79159fc936a3dc9c9d827c836f42d8219e472b3d8773d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Command: /usr/local/cuda-13.3/bin/ptxas -arch=sm_121f /tmp/audit-bw-min-sm_121f-ptx88.ptx -o /tmp/audit-bw-llvm-sm_121f.cubin. Output: 2,240 bytes, SHA-256 a09420d67fc46025b47f5ea0b60bd9c488060879e011b9e8c05eaa8f5debda21. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0cbff49774923817fcd3fe392ed98ce4839298f753be3a3a2045e2fbad51cae.
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6503456d0122b1eea2ff0ab89c97c0a581c085ac7a191aa64a709404b4ca9028.
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 af77c8ad395d7c645f9a605f6d1a87d3f02ac07bdc8a1c32dcb31fe3069d6088.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 566d8bcaab52298f01b20f07b20c7e36f8735d7f38e2ec771ec634d509a58fe7.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 561d11a944787fa985964fe99bde8cb2d6bc7894541a2d0dca73b945004f977f.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 49256692a73f54b955008d1fb880f317d6519b5f142e6fce4bda7ccfe63798c8.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f87c0a7c7fe3190edf0af7ad30c90d695b11040ff585357c3e5500650085431b.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b27fc629486b1be6e2f1672f3391f2cad48dcef3572e73dc2c95b6cd11a3e532.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e220cc23dece77c07832c94543b28e056a26d2d52ba1f95d24e36627caac760.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f380a3ff756634c1890814145efe5a3d576d276d3420f8eca0f5fa2e9bcbe63.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c99bffddeec90ca0a24db63f641d7b27006787d8b2c8a4cba23a54881528cee.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df72be0e5e2f976e0ae1a10768ed60640f34e4903a7afb9c641718c4afbf08d6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fef0a098f615d7a481f0a02fc3a3e93fda92c87f47d34dada08e32da26bf9057. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 717a76d554e553dcbc5676f0955ac259c99bcf966695307e807f22977f9c5ada. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5088d61dc4f383502beb5b4e75c86dbb62b0ee5cf6d8757a75d90679e521cd4f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0c48ea8630be568bade5e06bed37f6b71fb5515a5c15052313b0b4e5f749cb1f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b87519b1e54bed33ee379c81dec0dc4acc8ed83eae9f119007fa81815abed551. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6b854fc399ac96945e2207d11a7ecc8d14fed81b403f2d9856ed716ec79e3ed5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f05f6b168969ad56330e99c95467a52d622eeac0e19d593079194d37be69d6e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3edad0ec5c1cca330b876363b48b045e6982c95501db6a8c51f9a5a909269cc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12a37b4316c57c9f47e88b0b511f00b76e67f472036f5c1e8963529d6da81524. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM/ptxas/nvJitLink V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100a.ptx -arch=compute_100a; output 1,934 bytes, 12 exact heads, SHA-256 28553861b94c55284b1433b2e845af2105ced419c56b961e4400863503e79abd. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100a /tmp/audit-bw-lib-sm_100a.ltoir; output 4,216 bytes, SHA-256 38c17da342bf392a3b99f6300491c6fe0dabb5afbec7b5fe9fec20884ab463e1. Fresh PTX omits the API terminal NUL; the old 1,935-byte artifact matches its first 1,934 bytes. CUDA 13.3 rejects compute_101a and compute_101f as unsupported architecture options.
     - backend codegen on `compute_100f, ptx93`: `succeeded` — Fixture /tmp/blackwell-ldmatrix-libnvvm.ll is 2,456 bytes, SHA-256 20f944baaa677bed5e3cb3935ad108cd5994029650ef2e0f51568f9adda1b125. PTX command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/nvvm/lib64 /tmp/nvvm_compile_ptx FIXTURE /tmp/audit-bw-libnvvm-sm_100f.ptx -arch=compute_100f; output 1,934 bytes, 12 exact heads, SHA-256 37436e0ec8fc861b52dcad2feacc52fc2c3435e18eeb87ebbba4b029be215c9b. LTOIR command: CUDA_TOOLKIT_PATH=/usr/local/cuda-13.3 LIBNVVM_PATH=/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0 /tmp/compile_ltoir FIXTURE sm_100f /tmp/audit-bw-lib-sm_100f.ltoir; output 4,212 bytes, SHA-256 e1c62c7c5e47d3438046de138b654bb16a8e1c6d49fbf89fd2b75ddfb572d400.
@@ -9538,42 +9538,42 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - device link on `sm_121a, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121a -o /tmp/audit-bw-link-sm_121a.cubin /tmp/audit-bw-lib-sm_121a.ltoir. Output: 5,168 bytes, SHA-256 d7c7973b285796198f6a33547a82817c12d168558de213d435c53fb41587eefd. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — Command: LD_LIBRARY_PATH=/usr/local/cuda-13.3/targets/x86_64-linux/lib /tmp/elect_link_ltoir -arch=sm_121f -o /tmp/audit-bw-link-sm_121f.cubin /tmp/audit-bw-lib-sm_121f.ltoir. Output: 5,168 bytes, SHA-256 fb1e9dc5fd4c739c8efe7129ef00703d9099466905fcc0af613ec6c83080551c. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n8_x1_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_75`: `succeeded` — The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.
-    - backend codegen on `sm_75, ptx65`: `succeeded` — Emitted exact ldmatrix.sync.aligned.m8n8.x1.shared.b16 PTX.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_75`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d23e19589de35127bff359facfbc56c4c16a19fd1aeef8a4568c6080ad4c1a68.
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8eb444c30aad31f7ce8a70bcd06b91bae96ddd810e5c0e0bad67d669bba76463.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8a38549cf45825e7cf005005dcaa96cbdff09dd83f2b672a3d5bc23b8102018e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx65`: `succeeded` — CUDA 13.3 libNVVM accepted the exact shared-memory inline-PTX route.
     - device link on `sm_75, ptx65`: `succeeded` — CUDA 13.3 nvJitLink completed the combined six-form ldmatrix LTOIR route to a device binary. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n8_x1_trans_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_75`: `succeeded` — The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.
-    - backend codegen on `sm_75, ptx65`: `succeeded` — Emitted exact ldmatrix.sync.aligned.m8n8.x1.trans.shared.b16 PTX.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_75`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 078c3ec1a8767b2764c1a5dd2617806f672124e023a5c9fe5f3c40c6f5a36219.
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f448dfcfc8dc1e19e552c3f601db928004ab0f095c275cf846bc4e8adc3eceb.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a759f49ba62f6028179716e660082e3b36877cefbe37f9ad1aa91c27aebc22c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx65`: `succeeded` — CUDA 13.3 libNVVM accepted the exact shared-memory inline-PTX route.
     - device link on `sm_75, ptx65`: `succeeded` — CUDA 13.3 nvJitLink completed the combined six-form ldmatrix LTOIR route to a device binary. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n8_x2_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_75`: `succeeded` — The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.
-    - backend codegen on `sm_75, ptx65`: `succeeded` — Emitted exact ldmatrix.sync.aligned.m8n8.x2.shared.b16 PTX.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_75`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2228a4bf5e0bbaf7685dcbd5f0e890507df94d30aa9165cbe879037b84c74a4e.
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c86c69e86d0084dd76afccfa7dd76416f5a906eda8f245fd5eaadfb94eeedfea.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b42051503a9dcfd9cab9b052388ae403c55c7b81a776697ba93d8146a265b42a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx65`: `succeeded` — CUDA 13.3 libNVVM accepted the exact shared-memory inline-PTX route.
     - device link on `sm_75, ptx65`: `succeeded` — CUDA 13.3 nvJitLink completed the combined six-form ldmatrix LTOIR route to a device binary. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n8_x2_trans_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_75`: `succeeded` — The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.
-    - backend codegen on `sm_75, ptx65`: `succeeded` — Emitted exact ldmatrix.sync.aligned.m8n8.x2.trans.shared.b16 PTX.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_75`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b6c14f6dd0fdb52ad2e22873b19609ff47a1c8994bc5e3ce39c88bdc59a58f.
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51c2def02e859676a80a71edbbd6563d7dac75a59fb25899f3718a20705b267d.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67149c3d0b2f90b47e4e1f3f361f390adea9ffd4186e04bafaabd93b8091168f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx65`: `succeeded` — CUDA 13.3 libNVVM accepted the exact shared-memory inline-PTX route.
     - device link on `sm_75, ptx65`: `succeeded` — CUDA 13.3 nvJitLink completed the combined six-form ldmatrix LTOIR route to a device binary. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n8_x4_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_75`: `succeeded` — The recognized bare declaration materializes convergent, nocallback, memory(argmem: read), readonly, and captures(none) attributes.
-    - backend codegen on `sm_75, ptx65`: `succeeded` — Emitted ldmatrix.sync.aligned.m8n8.x4.shared.b16 with a four-register result struct.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted x4 PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_75`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 585528599fab5e664c529894e42a2b9604bc6e503f9eefc9d2117235a0fd35fd.
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3d0cc72dd11e3399dbfe72a912f451f36835e6c60ff54db70e1b6589b905dbbc.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c31b7c8cbef8fc5d164d732bfa611edc166ee875735791b5446e77f385f2f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `compute_75, compute_80, compute_86, compute_89, compute_90, compute_90a`: `succeeded` — CUDA 13.3 libNVVM accepts the declaration and produces LTOIR.
     - backend codegen on `compute_75, compute_80, compute_86, compute_89, compute_90, compute_90a`: `failed` — NVVM_ERROR_COMPILATION(9): unsupported operation.
@@ -9581,788 +9581,788 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - backend codegen on `sm_75, ptx65`: `succeeded` — This is the selected libNVVM lowering until typed-intrinsic support is uniform across supported targets.
     - device link on `sm_75, ptx65`: `succeeded` — CUDA 13.3 nvJitLink completed the selected inline-PTX route to a device binary. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `ldmatrix_m8n8_x4_trans_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_75`: `succeeded` — The recognized declaration retains convergent, nocallback, argument-read, readonly, and no-capture attributes.
-    - backend codegen on `sm_75, ptx65`: `succeeded` — Emitted exact ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16 PTX.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_75`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28e76c919b457758f8eeaf5f976e581db31431ae31ed1efbc30a050a31da1344.
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c74dd5efb5771ae005406a00b33baa93333dd3fbcaa2787d0dee90957c48871f.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5dbfbc0015f15b42f13d4bb3e15a049ffe769e096166c94bbb0aecfaa567d43. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx65`: `succeeded` — CUDA 13.3 libNVVM accepted the exact shared-memory inline-PTX route.
     - device link on `sm_75, ptx65`: `succeeded` — CUDA 13.3 nvJitLink completed the combined six-form ldmatrix LTOIR route to a device binary. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `lg2_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `lg2_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `map_shared_rank`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-cluster-memory` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM emitted both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 690-byte source had SHA-256 ad8a94dfa560d523bfe8cf963f8779132c960e1e6e5717e080e72819936f446a; the 1,007-byte PTX had SHA-256 005a7f90c8a3bd97a99792ce8a181d43c28fb318cd20dce6068fcb349800a951.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 3,712-byte cubin with SHA-256 1780aee43cd691bd7e3a79f4bc4b2d163e5cdfc11e08441ae318dfb091a5057a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-cluster-memory` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 418a46a6a0279c0e2c7391fcccae3b3bec1a660a3ffad28b1df2a7f63df297d6.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bf4f5d63cf41eae708cc3087c76cbea8b9812ba26f1c915646ef4b0205c51760. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-cluster-memory` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained both mapa instructions and one ld.shared::cluster.u32 with no extern calls. The 1,222-byte source had SHA-256 b966d8131b036df01f435a9dab89559917d5ac23d6e26d0adfc23a959d899cdc; the 2,208-byte LTOIR had SHA-256 a8006b43c0aacce7c48c04a1d3599127db090a21b4f2ed7cddc1d676a6243728; the 922-byte linked PTX had SHA-256 e8d9eea75eec5d6179fab2ba58d900c14bc5ed3af1c9ffe2c39f6e7c9fc798d0.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,840-byte cubin with SHA-256 5fc93334fcfa3fe77cee54dd6470bdecb562c8ecd205425b64df9d81a95ae091. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact max.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1bcd79e15bcc5f861613af0657054804f2cc98de0bc44082c84eeb6e06750d11.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 37e990ab010dfd7a6aa363c1bc12abfb5d4e121bceb7b89ac8056454e95a2385. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact max.bf16x2 route at compute_80 as part of the combined seven-form probe and produced 2,256-byte LTOIR (SHA-256 a9394a9ba0813c36f65bfc2636d7e80b27ab2ffcb6ded14a420d4127a824028d).
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,624-byte sm_80 cubin (SHA-256 1c8ff2eeb9e58f8b3defc22ed929cd48ec2e2f2beab30747f937a9a7869d74d3); cuobjdump showed HFMA2.BF16_V2, HFMA2.BF16_V2.RELU, HMNMX2.BF16_V2, sign/absolute HFMA2 forms, and F2FP.BF16.PACK_AB. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4070d164dfb7ec86737afcc51fdef8e9452193df799721221b988492eab3188.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9dc936add8025c51ff86d52cb72929dbbd1197519c0ffed33b53ba5eae61c7c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_80 as part of the combined three-form upper probe and produced 1,860-byte LTOIR (SHA-256 54983753f6c883e98ddef424042a704c0e38f32fcff6620ef12cf4734543d549). The instruction floor is PTX 7.0/sm_80.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 f70d50b154c66b2dc59909f29f449274a86d8300048982b8b1aa17b276193fe0). cuobjdump SASS contains HFMA2.MMA.RELU and both HMNMX2 predicate forms, matching the LLVM/ptxas path. A compute_75 LTOIR negative reached nvJitLink but failed terminal PTX compilation because all three forms require sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_nan_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_nan_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_nan_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_nan_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_nan_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_ftz_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_xorsign_abs_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_xorsign_abs_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_nan_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_relu_s16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 692-byte PTX module has SHA-256 8d5d0d27786cd1453843d71929ee823f7f864cf2bb31388b3f35924e5a78239b. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e5fc39021e5a2dd08e1b4c600bd65bfbd719061fd7182b4f87325793bcfeca3a.
 - `max_relu_s32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 676-byte PTX module has SHA-256 34e11547bce5d21764f671e36d47c30191aa492a57f253068f1a6f9ccb47d4cc. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 47ca916b21f4f81062a57a273ed3a2574d88dc7961f3ed061ee81135c02019a6.
 - `max_s16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 3a29f56af95dac7f78fa65c80cda474d1b2f71846240c18e28d1b227bed13bbe. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 019612a6c0a895437ac9e29439ca4404a44adf617072fb390270e41680557112.
 - `max_u16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 af376fac111934ef1cc375e57377b0eef4fd31e226db9eed52811fe89f924953. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 493b5e74e07a0232486d2df12ed0c80ed712745b36b51c17cfec678c28fdcb97.
 - `max_xorsign_abs_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_xorsign_abs_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `max_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_arrive`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact mbarrier.arrive.shared.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7df70d42d790f340f00c6f29da571b9b809bee9f1d093b742ce1a0e421272f46.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8d140364a2d31493758ea19f5e725bbfea20f14d6ce43489b079bb4789e428ce.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57bdb80f49dc0b09308813d01303ede4a2ba34e99e63aaecd6bea6813137fd7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the production barrier example to 4,616-byte LTOIR (SHA-256 63f7ec756a55d4fc813edda5424f2ec599969a2f96bca81c4260414a5f20049c).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,816-byte sm_80 cubin (SHA-256 7fc77da879df04e4ab109d1a74e5504ccb3e78afb2518eb6b1b74e06fb1dbcb5); cuobjdump and an isolated attribution kernel showed mbarrier.arrive as MEMBAR.ALL.CTA plus ATOMS.ARRIVE.64. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_arrive_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_arrive_expect_tx`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_arrive_expect_tx_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx86`: `succeeded` — The pinned Rust LLVM retained each exact cluster-scope instruction at the route floor.
-    - PTX assembly on `sm_90, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ba2b46453e1409c860af3f97180bd35dcdc2543d0ff333b4088e12af7b89ec97.
+    - PTX assembly on `sm_90, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 383a296ab6332c7761a49baa667e8ec2da174c6b80e39fadd82e44fb56a3602b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.6 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_arrive_no_complete`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the counted shared-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact mbarrier.arrive.noComplete.shared.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3eb1f951460f6bc67c6055899542860e1010e5db5406f5e2f9accee3cbc170c2.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 174b9c4a382274b14d72edd3fd22bdc8d2486f5cb87d6c8bb4a48e7bc66b7000.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b579f9834ca9bfb229788f02fc0ce145ba59fac61cbb61036bc8c552cb4bbb7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the production barrier no-complete regression to 7,056-byte LTOIR (SHA-256 3a7f7b31681ec0b1db3cf883967abd5e61a2d19449182bb83791ba1d94b38e42), retaining the dynamic count input and returned barrier state.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,704-byte sm_80 cubin (SHA-256 7193ea511d07aef4a88079f7b7fa606aa31820671c597477322dfd3c43d9e546); cuobjdump confirmed .text, .nv.info, .nv.constant0, and .nv.shared sections for barrier_no_complete_test, with the counted arrival lowered as ATOMS.ADD on the barrier state. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_init`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the shared-pointer declaration with convergent and nocallback attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact mbarrier.init.shared.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28fcac92a1ebd5fca5408ed34c7aa64ba3546bbe5714355a7ebfd14b92f760e3.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 50ba004fcfb85da96f6f3347966e4704dda4443ace6b57676924bd528c08bf0b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ab45efc26f5f14c513e6ab0ed925ffddbdfc1cb0e890ef4f445a40e88acebc9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the production barrier example to 4,616-byte LTOIR (SHA-256 63f7ec756a55d4fc813edda5424f2ec599969a2f96bca81c4260414a5f20049c).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,816-byte sm_80 cubin (SHA-256 7fc77da879df04e4ab109d1a74e5504ccb3e78afb2518eb6b1b74e06fb1dbcb5); cuobjdump and an isolated attribution kernel showed mbarrier.init as MOV/MOV plus STS.64. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_inval`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the declaration with convergent, nocallback, and argument-memory write attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend emitted exact mbarrier.inval.shared.b64 PTX.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76483bcaa272bf6c2c1688ff61ce630eba2cbc651af55d047b7048e43352adb7.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4effb270bc9d6ccd5d7c24a4518d532eb913541a234db702e0d50b7c4f7f0774.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 453f943cafeb60e5761ed5f0867ebb1e2519a3dd314eb9b0dc4b5e6ff6fbe65c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the production barrier example to 4,616-byte LTOIR (SHA-256 63f7ec756a55d4fc813edda5424f2ec599969a2f96bca81c4260414a5f20049c).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,816-byte sm_80 cubin (SHA-256 7fc77da879df04e4ab109d1a74e5504ccb3e78afb2518eb6b1b74e06fb1dbcb5); cuobjdump and an isolated attribution kernel showed mbarrier.inval as NOP. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_test_wait`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM backend retained exact mbarrier.test_wait.shared.b64 PTX and the predicate-to-i1 adapter.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6a81cac5ab0bf8f94c54775467e1df856112351608671f11a4e8e63ffb75f00.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df51ed0e79da4ce10d9298aa7e4fc3db4b75355044d51f95f92694acde7f9e50. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the production barrier example and retained the predicate-to-i32 adapter in 4,616-byte LTOIR (SHA-256 63f7ec756a55d4fc813edda5424f2ec599969a2f96bca81c4260414a5f20049c).
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 6,816-byte sm_80 cubin (SHA-256 7fc77da879df04e4ab109d1a74e5504ccb3e78afb2518eb6b1b74e06fb1dbcb5); cuobjdump and an isolated attribution kernel showed mbarrier.test_wait as LDS. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_try_wait`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM retained each exact instruction and predicate adapter at the route floor.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5dbfe0da451289673522a20ec37fd0664d5276bf7760b0d8a8484d613e82ca01.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24bb1e758b565ffc3df774b43f1cdc6f7b26a5f827d1add384e6ffcd78747080. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_try_wait_parity`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM retained each exact instruction and predicate adapter at the route floor.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5dbfe0da451289673522a20ec37fd0664d5276bf7760b0d8a8484d613e82ca01.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24bb1e758b565ffc3df774b43f1cdc6f7b26a5f827d1add384e6ffcd78747080. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mbarrier_try_wait_parity_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned Rust LLVM retained each exact mbarrier or fence instruction at the route floor.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted each generated module for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43181bf08ce9b3b668f72d19f5c2ff1fda31995dbff3b3e517bbe3ede29e620c.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2a64e0fd04a71a2ae3d940e37f22a332450115012ddf1d922d8598a15e5b3a0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 8.0 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the modules and retained each exact extended-mbarrier or async-proxy fence instruction.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into cubins for sm_90. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact min.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f59f875a1a686231cf0e0c09cd8ea1d7107aaf7ba547e54f0847ba4d71726b27.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51930c796a141b52e8382f91a54de007b9908945d7bdddf8085b3c9e4575557a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact min.bf16x2 route at compute_80 as part of the combined seven-form probe and produced 2,256-byte LTOIR (SHA-256 a9394a9ba0813c36f65bfc2636d7e80b27ab2ffcb6ded14a420d4127a824028d).
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,624-byte sm_80 cubin (SHA-256 1c8ff2eeb9e58f8b3defc22ed929cd48ec2e2f2beab30747f937a9a7869d74d3); cuobjdump showed HFMA2.BF16_V2, HFMA2.BF16_V2.RELU, HMNMX2.BF16_V2, sign/absolute HFMA2 forms, and F2FP.BF16.PACK_AB. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined PTX 7.0/sm_80 probe. Independent typed calls to the exact LLVM NVVM symbols emitted the same fma.rn.relu.f16x2, min.f16x2, and max.f16x2 instructions; sm_75 failed exact upper-group selection.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,163-byte PTX 7.0 module at sm_80 and produced a 3,496-byte cubin (SHA-256 5fa2890550ffe3793eadd967cc05c25b1bd9541c9a1404ef1ca2a67e7a9856f1). SASS contains HFMA2.MMA.RELU and the two HMNMX2 predicate forms for min/max. PTX 6.5 rejected all three as requiring 7.0; sm_75 rejected all three as requiring sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d54f22a9e49d8c8e53739fb3942a54749c50fe7749970cd67c4c77932fc1e11.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34e9c8ebf06b8ee9b1df5eda184d80fb12116914e533e5f11ecf26f337751abc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_80 as part of the combined three-form upper probe and produced 1,860-byte LTOIR (SHA-256 54983753f6c883e98ddef424042a704c0e38f32fcff6620ef12cf4734543d549). The instruction floor is PTX 7.0/sm_80.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,496-byte sm_80 cubin (SHA-256 f70d50b154c66b2dc59909f29f449274a86d8300048982b8b1aa17b276193fe0). cuobjdump SASS contains HFMA2.MMA.RELU and both HMNMX2 predicate forms, matching the LLVM/ptxas path. A compute_75 LTOIR negative reached nvJitLink but failed terminal PTX compilation because all three forms require sm_80. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_nan_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_nan_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_nan_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_nan_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_nan_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_ftz_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,256-byte bundle (SHA-256 ca9788d1820cbe1b33e96e93cde96bfbed43fd40e4fd3db00c89d926952de1e3) to 3,489 bytes (SHA-256 1fd153f40d985682ad06808043c0ff3f80df7beea2429e7e1b78734b2611e6a6). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 7,006-byte PTX (SHA-256 a3b8886623c8e4333c3e7137f558062fdad72ea41a266bace505fbc748cc1cfb).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 7,594-byte PTX 7.0 (SHA-256 7f929c917ce336149201aa549e34f586f85f44ec81d922ea0a9cc6150f8ee93d) lowered from the 1,651-byte fixture (SHA-256 7ac67205f3c8982a22b4539c936f875371b313cacdec884f6b86d4ea1568ecd5). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdae9fff36f881da4949ef1733359a4d7a42da97aef5b91c6fcc54110e92fdda.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b68f255f72c2dd2b53c1d108d4ad2a794232c88a1bed1f339530de647619684.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e31a11b0aeb2576b942df92106b620fad32ae20556d99c9aab2a19ba6f9e3d39.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b272c6ba1ff3228c9e5c9e329f7bc51d6849381f76306dc7da4b0e14537e5026. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 1,898-byte fixture (SHA-256 57018f57857ba7445a9dba481bdcf694fed97b079f1222a5a1cc027f29b7dd2e) to 5,669-byte PTX 9.3 (SHA-256 a787c4dde57c6ba45dee605a8bb9a641f42f95083695f847da4c3e16bde35d5b) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 3,400-byte LTOIR (SHA-256 9b77da4481d9db1b16f31b68a64d3e9eeb25480f8bc79e7c5b3b7c630f7074ff) under -gen-lto. nvJitLink produced a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,671-byte bundle (SHA-256 633c71128514c154f6265a81ecccaaf07403bb43f7506c43169745fa637186e0) to 3,657 bytes (SHA-256 cd5eff71081080c10456c5940a336f9cf2ab7aa81ddb9c4a74de48becf0e96e4). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted all 8 exact instruction heads once in 1,714-byte PTX (SHA-256 82bde5e15b99be603b82355420fe0833e33ac4fa99aa2913a948766a930cce5c).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_80 cubin (SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc retained all 8 exact, unique instruction heads in 1,799-byte PTX 7.0 (SHA-256 36fb6f2975e9fafed81eaac714d6aad034242d3fec23e3aaada22274bd340d10). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 5dc8854447fdad3cf21f7b63544880e98e1489646e4dda143b6775922c5fea09). PTX 6.5 and sm_75 comparisons failed with explicit PTX 7.0 and sm_80 requirements; stderr SHA-256 values are d319aa283a002fe31332d2bc659cd576cf12780ccef0da41902d951b45b72b7a and 51e54157842779ffd1528b1dd45f179d59f3bc8ec2f17d61689c4787ac250593. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26500126e694ac2e10fcb374209c66dbbbcad9ca536752588cd31449da50325a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ad13716b14d9f1c130dd26f18b2dde4a678235457a727ee829e9e61975bfe6a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f387222c7527cfd00ecb3febd31c4d322f7bb4eb26dcf0caa40bd6f5f477098.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3a7a09c07ca603d30285130db4c4d7dc1c005884300c37cb44a9de56faa9cc5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,403-byte fixture (SHA-256 f4484e08ab07cef192d8ea3dc4740a72500f5d024ea3bcf2baaefe776d0e5de7) to 1,731-byte PTX 9.3 (SHA-256 5413a5dfd3d56631ccd43e5434898b7e973ea5715520572a6d5c804eedb39a3c). The compute_80 fixture uses typed pointers for libNVVM compatibility. It retained all 8 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 3,752-byte sm_80 cubin (SHA-256 8e812aec859cf2aa948ef17cd6043a85c42b9a49eb1435059a2d0b27e6473853). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM emitted 2,392-byte LTOIR (SHA-256 48dc9720640e7eb97856090b58977d91371a2c2e2f85d49cbf404988cc31636f). nvJitLink produced a 3,752-byte cubin (SHA-256 e83464a0695696245147a9b239a9b8267fbb9b748d7a2fe6ed251282ca5ef984); its SASS is byte-identical to the direct ptxas SASS (SHA-256 1a6f737131188d03cb0ddb19d470085877276279827388288ea382a611f97ca6). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_xorsign_abs_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_xorsign_abs_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_nan_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_relu_s16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 692-byte PTX module has SHA-256 c10411ef8355e759ea266343683a7f30e971822dcdf3f47eb91c74bc44a92c95. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ccd77090590ed5077784a28895d4db44c0c3601913263b77b3afedf890a94db1.
 - `min_relu_s32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 676-byte PTX module has SHA-256 e835563e9676e88090632ce5c529840da1603ff6e65b3b88f2d722c626bd30d8. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05737e13d7ecd4d7d29377123087680401a96d0f6423842f9eb0c570c819a156.
 - `min_s16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 2916685ee0644fd8029821bab02c8cb4155b0ed1f3cce20b10c9dbe58edacf0a. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2b9b8cd02f5dbb4b21bbe79e50ef96e87c4ff8a61d113774a049c1df97e2d645.
 - `min_u16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the sm_90/+ptx80 candidate probe; the 652-byte PTX module has SHA-256 1d08c966ac1bea96a509823b5dc389325f2ac8a059d04619a8ff3e428e4f5d45. An independent generic-IR probe of the same backend selected this instruction from llvm.smin/llvm.smax patterns at sm_90/+ptx80, so the inline-PTX route matches the backend's own selection. No pinned CUDA 13.3 ptxas was available in this environment, so the terminal PTX-assembly stage is not recorded and the record stays at status "lowered".
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.0 / sm_90+: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bc038f0362d6863309339980bbb161649c4774ba6fc3e43058645c320efd1724.
 - `min_xorsign_abs_bf16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_xorsign_abs_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_xorsign_abs_f16`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 2,688-byte bundle (SHA-256 ae54a896b1439f56f9f5be8ec46209bee6ddec7f2e0687fd19290eba04fa9bc0) to 3,927 bytes (SHA-256 ffc1a0f2c931ce55db990064bfbd0855b8d47cb1de5e78285aa0e43506196623). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc emitted all 12 exact instruction heads in 8,226-byte PTX (SHA-256 a708a324123ae6698971b2f3f186bd7d439a0896cdab75d654fcf2af1a1a4f40).
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 12 exact instruction heads in 8,814-byte PTX 7.2 (SHA-256 5a17021f55c2bdac1ec99c2857a64379799f8eac704a4d8921ce4a0d0340906b) lowered from the 1,939-byte fixture (SHA-256 6495b61be3488c1de916e5604b49f090865b3d06bbbaca43f85da97dbdee4f6f). The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the typed-NVVM cubin, so both routes assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9891f48ef66b4b1adbba4b2d0d8ed77660dd0569f6a046166714f5c39c2dfe1.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 76e52e516ae7d1ce7c8785897b4d0eaca79b1817f277e15c08972141e9ef53aa.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91a82009fff395dc9c6629dbcfa55a280bfa1582b03cc556a52c5da08fd06474.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 12 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dddca11c0cd3da702ae90673fc678931937d5ecb6966625591655d806c94f87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 2,186-byte fixture (SHA-256 7b0e1b22096a297e428700d7be31bbd8e200b5ed0c2f9371928f2b4c40cb0a07) to 6,677-byte PTX 9.3 (SHA-256 09c57871571ba5317ee21cce0ee8cc05f3116b857084644a1923ca0d188a5524) retaining all 12 exact instruction heads.
     - PTX assembly on `compute_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df), byte-identical to the pinned-LLVM cubin for the same forms, so both front ends assemble to the same machine code. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,788-byte LTOIR (SHA-256 335c7c2f3488f1025b60376e0ec954b463cd8417b7acd62703748c9670b755b4) under -gen-lto. nvJitLink produced a 1,464-byte sm_86 cubin (SHA-256 3cf1096d7318aa9175bf174fe9254487a8d0030d98a961600a4e1d729739cffd), so the forms survive link-time optimization. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_xorsign_abs_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `min_xorsign_abs_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized the 6,304-byte bundle (SHA-256 0593b8e4f092e81c9b689784cc9fc4d546e5f2d0e4ae0d18de7942163ed77364) to 8,414 bytes (SHA-256 6df82bf883df8028630c87d191014cd929906625eac0f3fbc50f4f3520a83999). Every declaration gained the canonical nocallback, nofree, nosync, nounwind, speculatable, willreturn, and memory(none) attributes.
-    - backend codegen on `sm_86, ptx78`: `succeeded` — Pinned llc emitted all 20 exact instruction heads once in 2,986-byte PTX 7.8 (SHA-256 4d23b35b4fdbfed5e6ec50bcf6ad941c03997c98e383ef06e4434acaa0bedf9a).
-    - PTX assembly on `sm_86, ptx78`: `succeeded` — CUDA 13.3 ptxas produced a 1,464-byte sm_86 cubin (SHA-256 ac87a5754b27655c80d8e61387ce051754d199d422f7d209f16a4208816152df). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_86, ptx72`: `succeeded` — Pinned llc retained all 20 exact, unique instruction heads in 3,529-byte PTX 7.8 (SHA-256 a0c8374e49cae0b20a1f2b2a9f5999ec9c69f3b7b023a4661e05d9ca02f47867). cuda-oxide maps catalog PTX 7.2 through 7.8 to LLVM's available +ptx78 feature. The typed and inline instruction indexes match.
-    - PTX assembly on `sm_86, ptx72`: `succeeded` — CUDA 13.3 ptxas accepted the 3,529-byte PTX 7.2 module (SHA-256 7d8f296b827a63d0fae7424822bcbe9df8758cf8c1c7a82016b7d501a9c669ab) and produced a 4,136-byte sm_86 cubin (SHA-256 dd975a09af37a8a7a825d287383ebb15f492f4d004223b494af0c0ba1185c26d). PTX 7.1 and sm_80 comparisons failed with explicit PTX 7.2 and sm_86 requirements; stderr SHA-256 values are bda2abd9758354188439be0cb81ea5693be653c095aacbd153848d0a3df46a0c and 55636d5221bf3d6b9eca0222d13f7957936a20d1bfe09e491f2756a6eb234afd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.2 / sm_86+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 27e9f1f0758625507ee908c70a275f9cb7447bcdb75b0822fa910ee772139b4f.
+    - backend codegen on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b81819f4a18559a1a6f561616e50185c2abd3ea4a2c8a0f1ab0dc6fe2e5cafec.
+    - PTX assembly on `sm_86, ptx78`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0b8c52cece4dbd29ab86d90ed6514eb59335ce27cdb377ca2ec34f53fb913c4b.
+    - PTX assembly on `sm_86, ptx72`: `succeeded` — Revalidated 20 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f61ead0f078446de8e8a76826a0ab1bb3e77fddd24828294524d684b3c67fb7e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-extended-minmax` at PTX 7.2 / sm_86+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_86, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5,308-byte fixture (SHA-256 c6d1fa1cea32871917feebe90218b1adfe95fa718bb6ead6f0e4068db9058c09) to 3,534-byte PTX 9.3 (SHA-256 2b11bd322b46f9631b9e17aae19d1268599b99d2d9cdedddc3fd3d4bb3a1bc72). The compute_86 fixture uses typed pointers for libNVVM compatibility. It retained all 20 exact, unique instruction heads with no extern or unresolved LLVM calls.
     - PTX assembly on `sm_86, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 4,136-byte sm_86 cubin (SHA-256 389a4f01ae0378d95033280418358bf93e98345a4f7270dd8fdc87cf78dcc5ec). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_86, ptx93`: `succeeded` — libNVVM emitted 3,768-byte LTOIR (SHA-256 b9418d3902ff675b1affdc564d81f75455f0a5195a87363266eef55da96fa61f). nvJitLink produced a 4,136-byte cubin (SHA-256 487bf670602b607f9609ace78511744c63e7264d3e5a8ff5a966330842b67f2f); its SASS is byte-identical to the direct ptxas SASS (SHA-256 c3cd354972834542b2adac632c8d9839974d8e64c9c5b255bb150a556472d3c7). Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k128_s32_b1_and_popc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0786eb2ac48a149064adcb908b142249e47b70b7dc6b77221c691534d54f8363.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d7868cbe80c20cad2dafd27dc6429792a50fd3b4efb0da2dd55b4d836fedaef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 165,143-byte compute_80 NVVM IR with SHA-256 50e40ba550994145a6c57741b22786e09252da4d786c0e296d8de14cfe44d30e and 47,744-byte LTOIR with SHA-256 060d4347bb268f38e14ebe8c5ad9db64fbf1781d578f14c8799f3d8f54f2db71.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 20,456-byte cubin with SHA-256 aa2ce30ecae34abe738c760e7d3f2b20d35744652afdae0b6abb29b4bf10d8a8; cuobjdump contains all six distinct BMMA shape and AND/XOR heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k128_s32_b1_xor_popc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_80 floors. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4fc3a76fa4f4f3118a8078863963902f0b33ebeecb269fccefaf2d785a9171f4.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 66efb066dd43977add4c7b6812754fc0affb5166a6dee22b060759d4eac2e5ab. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 165,143-byte compute_80 NVVM IR with SHA-256 50e40ba550994145a6c57741b22786e09252da4d786c0e296d8de14cfe44d30e and 47,744-byte LTOIR with SHA-256 060d4347bb268f38e14ebe8c5ad9db64fbf1781d578f14c8799f3d8f54f2db71.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 20,456-byte cubin with SHA-256 aa2ce30ecae34abe738c760e7d3f2b20d35744652afdae0b6abb29b4bf10d8a8; cuobjdump contains all six distinct BMMA shape and AND/XOR heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_f16_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - PTX assembly on `sm_75, ptx70`: `failed` — ptxas rejected the form below sm_80; the diagnostic SHA-256 is f6582289bf7df33af5bf1c016ce0f73335a4a37ca257c42d201fe9cfefbe134c.
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - PTX assembly on `sm_75, ptx70`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the form below sm_80; the diagnostic SHA-256 is f6582289bf7df33af5bf1c016ce0f73335a4a37ca257c42d201fe9cfefbe134c.
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5775-byte typed-pointer IR fixture with SHA-256 42a32e7fc0a6d40513480eff6d1fef5603bd2dc07621b1a1e3d8d946dac1f8e6. It emitted all five forms once in 6449-byte PTX with SHA-256 b7d87c4199fa520ab749fe1dd8f3b37a3b6685c9a7a9e1c8cebbf99342d2bd50 and 3676-byte LTOIR with SHA-256 582cd60f2d911bc38214437d619a334cb8e5cb96d6ec811329866aea4d027855.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced an 11048-byte cubin with SHA-256 e89808e3781dfa9b928f239cf2943777417f6ad789fb6c10f6981777aff6d4ff. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced an 11048-byte cubin with SHA-256 1d242578ce7378f849df0dbcd28bd43a60af64b7f76359b7e5510ecd23a14572. Its 52dc9bf332bdb595b2910550d95f0d0f8c44477440c52977bcc6b0699cecffcb SASS dump contains exactly HMMA.1684.F32.TF32, HMMA.1688.F16, HMMA.1688.F32.BF16, HMMA.1688.F32, and HMMA.16816.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_f32_bf16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact BF16 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0a3bd41a37b501ef0419e036d4a2325f28dee3d5d47c050ab1af6d52b9904385.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b0d9dbc95cc97fb422f1b7b9e39d1d68c866b4ffcf48c9a32f8fb917454bb82. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics example produced 11,096-byte LTOIR with SHA-256 71862b5aa53785a7f7ef9f7d86ddc5a4b88d893d6408ec2532f1f0a230173fda.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 15,464-byte cubin with SHA-256 955a6f99c580ad85d3c30a003f34f71ff9ccdd595386880f524daea4f1745eba; SASS contains HMMA.16816.F32.BF16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_f32_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact F16 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cb21825affcdfcb250dea8bfead8827d200376f1b6dfb69b7659d670b4fe0a8d.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d25700d38dead0336c44add15c4dc07020aab017e285b96e0e94bf091e743e4e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics example produced 11,096-byte LTOIR with SHA-256 71862b5aa53785a7f7ef9f7d86ddc5a4b88d893d6408ec2532f1f0a230173fda.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 15,464-byte cubin with SHA-256 955a6f99c580ad85d3c30a003f34f71ff9ccdd595386880f524daea4f1745eba; SASS contains HMMA.16816.F32. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_fp8_f16_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM compiled a 13717-byte typed-pointer IR fixture (SHA-256 add55fcb64ff5a318dc20e775f57cd17d775dac1215a43828a3b2d1b6b513cd6) to 9034-byte PTX (SHA-256 902cb73837cc693df499fd107cc437dd344315e57e4b29ad8b980d269452f0dc) and 6932-byte LTOIR (SHA-256 ff6619b9b8cf44f6bec07aeca7d6f2ed5bffda01501d0d0a96a06bf4fdd48f2b), with all 16 forms live. An eight-call convergent duplicate fixture remained eight instructions. The same fixture emitted all forms for compute_90 and also emitted compute_87 intermediates that terminal tools rejected.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas produced a 10088-byte cubin (SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc); its 43177-byte SASS dump (SHA-256 2693f93687940907f276c0383cbdb32e17a2b4d7a94886cc14895fea9e5d2f6b) contains 16 QMMA instructions. The same all-form PTX assembled for newer sm_90 into a 13296-byte cubin. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10370,12 +10370,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected all forms because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is d25b354f14d618d6202a98bbfe3292fb19e1a0894aea8f5aba28d0e2530a071f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is bf2ed0ca6d60259bc52c0449bbbf7a68af6811b581e338d04fdeda562b6587fc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f16_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM compiled a 13717-byte typed-pointer IR fixture (SHA-256 add55fcb64ff5a318dc20e775f57cd17d775dac1215a43828a3b2d1b6b513cd6) to 9034-byte PTX (SHA-256 902cb73837cc693df499fd107cc437dd344315e57e4b29ad8b980d269452f0dc) and 6932-byte LTOIR (SHA-256 ff6619b9b8cf44f6bec07aeca7d6f2ed5bffda01501d0d0a96a06bf4fdd48f2b), with all 16 forms live. An eight-call convergent duplicate fixture remained eight instructions. The same fixture emitted all forms for compute_90 and also emitted compute_87 intermediates that terminal tools rejected.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas produced a 10088-byte cubin (SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc); its 43177-byte SASS dump (SHA-256 2693f93687940907f276c0383cbdb32e17a2b4d7a94886cc14895fea9e5d2f6b) contains 16 QMMA instructions. The same all-form PTX assembled for newer sm_90 into a 13296-byte cubin. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10383,12 +10383,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected all forms because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is d25b354f14d618d6202a98bbfe3292fb19e1a0894aea8f5aba28d0e2530a071f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is bf2ed0ca6d60259bc52c0449bbbf7a68af6811b581e338d04fdeda562b6587fc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f16_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM compiled a 13717-byte typed-pointer IR fixture (SHA-256 add55fcb64ff5a318dc20e775f57cd17d775dac1215a43828a3b2d1b6b513cd6) to 9034-byte PTX (SHA-256 902cb73837cc693df499fd107cc437dd344315e57e4b29ad8b980d269452f0dc) and 6932-byte LTOIR (SHA-256 ff6619b9b8cf44f6bec07aeca7d6f2ed5bffda01501d0d0a96a06bf4fdd48f2b), with all 16 forms live. An eight-call convergent duplicate fixture remained eight instructions. The same fixture emitted all forms for compute_90 and also emitted compute_87 intermediates that terminal tools rejected.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas produced a 10088-byte cubin (SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc); its 43177-byte SASS dump (SHA-256 2693f93687940907f276c0383cbdb32e17a2b4d7a94886cc14895fea9e5d2f6b) contains 16 QMMA instructions. The same all-form PTX assembled for newer sm_90 into a 13296-byte cubin. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10396,12 +10396,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected all forms because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is d25b354f14d618d6202a98bbfe3292fb19e1a0894aea8f5aba28d0e2530a071f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is bf2ed0ca6d60259bc52c0449bbbf7a68af6811b581e338d04fdeda562b6587fc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f16_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM verified all 16 declarations and canonicalized them to nocallback nounwind memory(none), without convergent. The 14810-byte source and 15723-byte canonical IR have SHA-256 06c1e45eb81825111c318a3a168a39c51ec2367ec298170aa33c68913c79e18c and 2d6806628e2e041fbd723990a0bcca1de0a95b688984fcefe937e3e602f2fc7d. O2 merged eight representative typed calls to four, while the convergent sideeffect inline fixture retained all eight calls. The typed CSE source/output hashes are 8875c8c486b6e07e1c29ad6e3f050a42b49d7d670ea16d36b1cc24b2e9961500 and 5dadcecbdff3a4fcc2d1eed72ca2b066c8b541aedaad69df839a5c00ce38137e; inline hashes are afe744162d4cfb8d06393b39d65d16354d5af131e3e75f13a8891a361be6e23c and 4574a453fe7ef2d8d44d2cbe7973a02a31bb40459a2461f22c924d86731a13e8.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — Pinned LLVM emitted all 16 typed forms in 8609-byte PTX (SHA-256 69d1ed4f97813f2410219952455bd8f0d17930290be1f129edf88c0e270f7506) and all 16 convergent inline forms in 9234-byte PTX (SHA-256 46bdcf6be3a1c0811baca29a4ac3c3b421216c03b33376ffb9fa97e304479329).
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas accepted all four k16/F16 forms at PTX 8.7. The PTX and 3496-byte cubin SHA-256 values are bb964ef8b6262d4c55d102c8660ffde741c833c50669fda6040edb754792876a and 0ec074dc7c5ed05604bebdd41e483f8a636c32bb90d8e6683cc74e7a163a9e63. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede12485cbc7a4357e5e12b1f25d58314ccad2e9479bd406bf2d9121acdd9d8e.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fba2ec2311b639db327e9b0f750c7919c968fbdbea64ea28a198893b5c797b61.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0014c678f9bffe285cd97bcd41156276b1f1c52180259c18ca3a79864beef7db. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected every k16 form below PTX 8.7; the diagnostic SHA-256 is 36ca3350917752b188d197cf2e28d18d183be3dd01c12ea3847e9c5bb82f15af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM's typed selector and CUDA ptxas both rejected the family below sm_89. The all-form ptxas diagnostic SHA-256 is 83ee414277eb4c4fb9afa8cce61ec828379167b192c004713d7b214d589a5069. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM compiled a 13717-byte typed-pointer IR fixture (SHA-256 add55fcb64ff5a318dc20e775f57cd17d775dac1215a43828a3b2d1b6b513cd6) to 9034-byte PTX (SHA-256 902cb73837cc693df499fd107cc437dd344315e57e4b29ad8b980d269452f0dc) and 6932-byte LTOIR (SHA-256 ff6619b9b8cf44f6bec07aeca7d6f2ed5bffda01501d0d0a96a06bf4fdd48f2b), with all 16 forms live. An eight-call convergent duplicate fixture remained eight instructions. The same fixture emitted all forms for compute_90 and also emitted compute_87 intermediates that terminal tools rejected.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas produced a 10088-byte cubin (SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc); its 43177-byte SASS dump (SHA-256 2693f93687940907f276c0383cbdb32e17a2b4d7a94886cc14895fea9e5d2f6b) contains 16 QMMA instructions. The same all-form PTX assembled for newer sm_90 into a 13296-byte cubin. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10409,12 +10409,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected all forms because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is d25b354f14d618d6202a98bbfe3292fb19e1a0894aea8f5aba28d0e2530a071f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher; the diagnostic SHA-256 is bf2ed0ca6d60259bc52c0449bbbf7a68af6811b581e338d04fdeda562b6587fc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f32_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — The all-form libNVVM fixture emitted all four k16/F32 forms in PTX and LTOIR at compute_89, retained representative duplicate convergent calls, and also accepted compute_90.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — The 16-form ptxas cubin is 10088 bytes with SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10422,12 +10422,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — libNVVM emitted compute_87 intermediates, but ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f32_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — The all-form libNVVM fixture emitted all four k16/F32 forms in PTX and LTOIR at compute_89, retained representative duplicate convergent calls, and also accepted compute_90.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — The 16-form ptxas cubin is 10088 bytes with SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10435,12 +10435,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — libNVVM emitted compute_87 intermediates, but ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f32_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — The all-form libNVVM fixture emitted all four k16/F32 forms in PTX and LTOIR at compute_89, retained representative duplicate convergent calls, and also accepted compute_90.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — The 16-form ptxas cubin is 10088 bytes with SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10448,12 +10448,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — libNVVM emitted compute_87 intermediates, but ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_fp8_f32_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — The all-form canonical fixture proves the four k16/F32 declarations and the omission of convergent; representative typed calls CSE while convergent inline calls do not.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k16/F32 forms in 2435-byte PTX with SHA-256 51bce9ab0dc29633e0c3a1f706f81e53237fc7dfd11c720798ebd443d850d30e.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3752-byte cubin with SHA-256 1263567f5478c0d06f579c67cb435d95236f5ad1cc08861312e0d4dbb2f14641. The aggregate 16-form fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 400659013a48be7da03f9e9c28b77b573f44005ae068522e5c93b9b178041746.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2e492028849ccd4dca87b4ca16af886692ba8f12542761bdb8e1e7fab7c440d8.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6df68d45c08985dcb2ead31bfbf4cf27f62195f7a0105bdbf3ae9d9066566cd2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k16 FP8 MMA below PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — The all-form libNVVM fixture emitted all four k16/F32 forms in PTX and LTOIR at compute_89, retained representative duplicate convergent calls, and also accepted compute_90.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — The 16-form ptxas cubin is 10088 bytes with SHA-256 bb4610329a1066be83c4306c9b46b5f58300387b496a33415420537be89e4cbc; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -10461,94 +10461,94 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — libNVVM emitted compute_87 intermediates, but ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected compute_87 LTOIR because standard FP8 MMA requires sm_89 or higher. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k16_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping s8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a48e36d7da44b51913001f1eb7d4d1a4c3fe187828db3a37451079cfb79bd041.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3f2318b8382d87d1f67a2cc34a6842dbb8d910a85130f0d3d991690b39d6f1c2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite s8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32f364dd111daf1d85405149c434903c9ba832ee2d5d6e3067ab468e61bcd713.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f136aaf2137b3c1f962fdcdbeaf7f7ed4ad688be4fb5cd55dadaf7ac100fc153. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping s8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c7ad5689bb1dc7d9f6e66a3b7024246cecd620c8499b9468999c5efc7cf4ef38.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2318600ab58b33347303547db32bfbfdccea5210d1401891255d10c18a8d89cd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite s8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 516ff5a0c66113c514e59c0272876c73360c509581521ac8ef9c6c7b1a02c82e.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b15a9cd586a5fbde549415441be60eae06e57d99604ba2350ca476c0066f6a1a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping u8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 db2d33329053739e1f39831d56e2d27e487dd334679bb09de2f5c8f80eb1de65.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0a674a423572b924a7f2e6dd0a2c69d601e5d2c4682dfa1414103e7045a686c6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping u8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9921a009519c555a2adb72baf0e136405ef6b850b9a19d83f10e7d0b3d763f71.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 969755d83267f71bbb72953caff716f8eae65ee1d14c4a293879a0a21338c337. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite u8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab5b0e911466d87855d48864c54ff2da53c2e4e5b940b1c0812b378ea453e585.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a300eed5eea4deeaf14e77ce212087a11d0a912a110956a8f98d7ef0219018cf. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k16_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite u8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 332115a18b93fa8d2c7916c9fe028636dde30234d0b197244146f55eca052148.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8301b6cd488dea73b3278090a9a30910809f2afba5c3e9b58b21a1d84cdca8bd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k256_s32_b1_and_popc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b08bb4ac9ef5d95f902982d6d6e4da9bc6e758f02a86e4f0da412b6dd22f561c.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd92d37a566e4ee028ef24ef3bc175b98a0ade64ebf874e29c45a409119ee2bb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 165,143-byte compute_80 NVVM IR with SHA-256 50e40ba550994145a6c57741b22786e09252da4d786c0e296d8de14cfe44d30e and 47,744-byte LTOIR with SHA-256 060d4347bb268f38e14ebe8c5ad9db64fbf1781d578f14c8799f3d8f54f2db71.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 20,456-byte cubin with SHA-256 aa2ce30ecae34abe738c760e7d3f2b20d35744652afdae0b6abb29b4bf10d8a8; cuobjdump contains all six distinct BMMA shape and AND/XOR heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k256_s32_b1_xor_popc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_80 floors. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e760db526d0a3aef6cdfb3a2032edb8ef925f23d7c046b15eb455e304c0a66b2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f577f397dc588d57cf9709716a198f6b17f1d4f6fa071c808bcaff2ddb58b8dc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 165,143-byte compute_80 NVVM IR with SHA-256 50e40ba550994145a6c57741b22786e09252da4d786c0e296d8de14cfe44d30e and 47,744-byte LTOIR with SHA-256 060d4347bb268f38e14ebe8c5ad9db64fbf1781d578f14c8799f3d8f54f2db71.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 20,456-byte cubin with SHA-256 aa2ce30ecae34abe738c760e7d3f2b20d35744652afdae0b6abb29b4bf10d8a8; cuobjdump contains all six distinct BMMA shape and AND/XOR heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m1_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10564,24 +10564,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m1_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10597,24 +10597,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m1_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10630,24 +10630,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m1_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10663,24 +10663,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m1_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10696,24 +10696,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m3_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10729,24 +10729,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m3_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10762,24 +10762,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m3_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10795,24 +10795,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10828,24 +10828,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e2m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10861,24 +10861,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e3m2_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10894,24 +10894,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e3m2_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10927,24 +10927,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e3m2_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10960,24 +10960,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e3m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -10993,24 +10993,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e3m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11026,24 +11026,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e4m3_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11059,24 +11059,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e4m3_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11092,24 +11092,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e4m3_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11125,24 +11125,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11158,24 +11158,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11191,24 +11191,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e5m2_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11224,24 +11224,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e5m2_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11257,24 +11257,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e5m2_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11290,24 +11290,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11323,24 +11323,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f16_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized all 25 typed declarations to nocallback nounwind memory(none), without convergent. The 20899-byte source and 22444-byte canonical IR have SHA-256 3b238c50ffde36e35e6b958afe7dd843752be7bda45273e168fff2c54cb49cd1 and c8658be15eaf41af66a4daaeb0b9f7c1b817be9b18580b1fac7ebcb2b3cef711.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms in 6609-byte PTX (SHA-256 727bef86d8a68a50f5b96a3f84469e7185ea8add6dd1b0dff7cd7dbd45d6aabc), but O2 merged two identical calls into one. The selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 e693bbe87804706b76cdf9902eaf71a7a48022f64f7d2c705890ad62fd370ff2) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 7a1bf2ee5a48794bd9d997c9ac499d06f76c05671eb78fd1dc64fd722e2e1b64); its 24812-byte SASS dump (SHA-256 a9e7f9947fb88d2c61dde81c63c97735ffff55e9ea002f16fbf5043b971bd7d4) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 3ca1691177595b79cbfddd54c68abc1a9cc0f4a357d2cd276d882b57c1b21c64) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 07cb35ea5c125c9930da79d8e61623134eaa49cd7bd9a5b2ebf19392fb345d33); its 24789-byte SASS dump (SHA-256 757d6d29c8900288ef5a5a167f2ac0967a7b9f1dcdb9dde326a0e7f96c3cd2be) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 cc5d67077f66e2c9f83500b817ee7a85830d81eb6ad938abd61946dbfe2d1457) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 46616ac6f0caf7dd3720425f8bed3f501df00a9e74b25c26a8aab65f1d798a5c); its 24812-byte SASS dump (SHA-256 7a820d3de6a6fd1789afe3e29d2c8ced21d2f38b696ee701dfdf4f1c942a880d) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 7649-byte PTX (SHA-256 871086b1542b0725ee84ab90c5aedcde58fe7f027e98682011aef6f3c7fd376a) with exactly 25 unique forms and all 50 result words live.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddaea9ba0f55e96ba5ca75ac80ca4a6cd826b79eec9ee6bd1f92903a4074c6c.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a433322fc353c7b9ebcc4a6d361dcc9cf88f317cd9850508318da36abb0e91e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4367c5eedef01ac6907cedc4a0125d1b3c8f8826f55b0326b3f8f6c55e1cb65b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1efc6b229a0bd99382a2d571a6faa544664418253c7188db55b153b9450e66d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ffb977394436acd64442168bef5ce9a88b21fe85eb24642791b67da14b0f8c3.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 89b7a03a7da15753cfc9812412fdc50cd30581d2c4fc007f19e196421cec98a8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47f54eec937ce0473aa982c2506eb89ce12a8af530283ac1b448ec6d7c1639e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a05245e93abfb751413bb38b13ec602e8dc4d0229e8ea6cf8c6db7f45b2aeccc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7a5af3bee9337495f57f5788eff113d87311df57a732cd80d8e566cd8ff59de.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40bf3d5137031aaf4db7dcf8c5213e1d2fe97a556086b2c1fb2d8f2cf1d39c2d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted all 25 typed forms in 6512-byte PTX (SHA-256 f3ae18918d3e7f9e0bec9c61e2e1311480fba0c25b761c8e6029e971089af51f) and 7528-byte LTOIR (SHA-256 0088b1e372a1e11616976865a952786ee045fd9dc761b1944a2823795b5748d3). For a 1310-byte duplicate-call fixture (SHA-256 df619d51a8b6e43fbe33f341608739f534798ba10daaac71e62182c06e3d7556), it merged two identical calls into one instruction in 1297-byte PTX (SHA-256 262c703198f1ee28ab23f8440f0e19de21d6e74ddf6c2d5c35f371a19cd77d9a) and emitted 2412-byte LTOIR (SHA-256 6a2ef1e2cbb89c86b9aed5d32109fab85b3bf7ee76d7e7ebcc56c7d756f3ff84). The selected convergent inline route retained both calls.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 7544-byte PTX (SHA-256 655e1ff713860e19aaf27877cc7f2063730a6151e82ba3308cf5366f8adef73e) and 9492-byte LTOIR (SHA-256 3ed7fbb9cf41237bd689e6b770fea485fb1a1fdfa9bfcd20b4d33eeaa35c10a8) with exactly 25 unique forms and all 50 result words live.
@@ -11356,24 +11356,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 9600-byte cubin (SHA-256 729e133ea50cd435fa9d347bcacb58dcd48892f2befbd8e6e99ae448f0d89043); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 9600-byte cubin (SHA-256 0d6f9b1666e1096c7e7bcf6c0e6dfcc6412ed532a01204de4164731c7d9e4fda); its 24789-byte SASS dump (SHA-256 86c6f43d1c884935d0fbcef2a2cdbd86e163c6c7d2288b86a2ca2fa653013a58) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m1_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11389,24 +11389,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m1_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11422,24 +11422,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m1_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11455,24 +11455,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m1_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11488,24 +11488,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m1_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11521,24 +11521,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m3_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11554,24 +11554,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m3_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11587,24 +11587,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m3_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11620,24 +11620,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11653,24 +11653,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e2m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11686,24 +11686,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e3m2_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11719,24 +11719,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e3m2_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11752,24 +11752,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e3m2_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11785,24 +11785,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e3m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11818,24 +11818,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e3m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11851,24 +11851,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e4m3_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11884,24 +11884,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e4m3_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11917,24 +11917,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e4m3_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11950,24 +11950,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -11983,24 +11983,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -12016,24 +12016,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e5m2_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -12049,24 +12049,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e5m2_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -12082,24 +12082,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e5m2_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -12115,24 +12115,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -12148,24 +12148,24 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_f32_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — LLVM canonicalized every typed declaration to nocallback nounwind memory(none), without convergent.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Typed lowering emitted all 25 forms, but O2 merged two identical calls into one; the selected convergent inline route retained both.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 170419484989ac4bd991bf8cf52ced9a47d3c537bd31c1c9b82ad9d34d8519e5) with exactly 25 unique forms.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 91b1db2dc0f4505badc8dc216f4d766f883046f2b51877846cdb4f304d291689); its 32140-byte SASS dump (SHA-256 3943ff27ac8050cca484e822fd0605e465768a53c60c18e70ad5e150c63c6760) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 7ac34b61b50100405a6eda6efa66638714d3d62d0a9489e3d8d94b29d4e3e187) with exactly 25 unique forms.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 cc9ca297ca74ba08abf6417dab5cda3cfaae2ead91bab33aa0db5cb91436c879); its 32117-byte SASS dump (SHA-256 7fed82b233b7be9f248a25ff19b9f63a7d9ee9bf1c1fa9fbe1172dc3e19988c0) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 dfc49a74b206a05bb12dc61f2fa1ac1b172abb8a43a3ac772305f82c2ae5fb1f) with exactly 25 unique forms.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 eeb6a6169d0f2ea9c74083e091fa459c459399c79903533ad71b177e30b028f6); its 32140-byte SASS dump (SHA-256 9185dde3c15bd63a879d2f697670884c5cabdd75cb7417ea60bbe37d8e66638c) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 10665-byte PTX (SHA-256 71265a2f5612e88e43369992da9758934c11a01f488317d05afca5b83a12ba8a) with exactly 25 unique forms.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100a, ptx86`: `failed` — ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_110a, ptx90`: `failed` — ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120, ptx87`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121, ptx88`: `failed` — ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_120f, ptx87`: `failed` — ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121a, ptx87`: `failed` — ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_121f, ptx87`: `failed` — ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dec66bbf6473f68e7a59a33220f594a90fb34d05f385d227193a4db4f3420f0d.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a3edd89f7df0ae1318cf5f137e1d3e3b321b75454b4d4b99cf7d7710e91bc3f.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 21c16f36f9dc183c4ae0bcfa0ccfd9bca7fd3330c2b7c51b8a595313c970043c.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d58dfaa643cc09116bc46dc887be5e72f28ebcc54d3044a46c847030a2aa87fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f6fe6fd270310333687ed32ca5db1d611c071bf8dbdc9b135eee69fc28a256e.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4db46d762e84307ab30f4fa736905e1fd004458c1a4cbb4fbfc52705038400fa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f5c4721d1daebeaddd4890108ee6c5da2d5503f8564a4adebd77a1e9ab6f3451.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0e959132d401a1526fe4bfa147a563bfd725fef56ae4ad7ebb251ad957735e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8257f1ea87a21487c98b11f3429a45c9fe6e2f4e35d77b45ea8de269aa45e4ab.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a8b866775acd390c035007d5c96fb73f10532a2f749c15acfda847220a698274. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected PTX 8.6 because FP6/FP4 MMA requires PTX 8.7 or later and is unsupported on sm_100a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_110a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the instruction because FP6/FP4 MMA is unsupported on sm_110a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_120. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121, ptx88`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected kind::f8f6f4 on unsuffixed sm_121. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_120f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_120f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121a because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_121f, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected sm_121f because PTX 8.7 does not support this target spelling. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM recognized the typed form but merged two identical calls into one; the selected convergent inline route retained both.
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM emitted 10560-byte PTX (SHA-256 fd7f653c260b48292a2d6ff8b3654724e86a5ef6cba27d89ef1af0d7dd06b65f) and 12096-byte LTOIR (SHA-256 cf491e7bc314c99993da01f489a35b897ea35320b722e9cb1c7776b0993ea5c0) with exactly 25 unique forms.
@@ -12181,12 +12181,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — ptxas produced a 11760-byte cubin (SHA-256 42a912f7444829111d55b9d096e3f65f996e03da85f9b29624b6623c2eeb3449); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced a 11760-byte cubin (SHA-256 314af8f3e609e09a7ee6713b441bba1e80b9d73d346300c3d837525b6b146766); its 32117-byte SASS dump (SHA-256 9afa25b36cbfef928bbf799565cb9548824ad5d16290088e35462dee6b6aab7b) contains exactly 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_fp8_f16_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F16 forms at compute_89 and compute_90. Admission remains PTX 8.7 because CUDA ptxas rejects this accumulator at PTX 8.6 despite LLVM's PTX 8.4 selector predicate.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F16 forms in the 10088-byte all-form cubin; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12194,12 +12194,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f16_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F16 forms at compute_89 and compute_90. Admission remains PTX 8.7 because CUDA ptxas rejects this accumulator at PTX 8.6 despite LLVM's PTX 8.4 selector predicate.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F16 forms in the 10088-byte all-form cubin; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12207,12 +12207,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f16_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F16 forms at compute_89 and compute_90. Admission remains PTX 8.7 because CUDA ptxas rejects this accumulator at PTX 8.6 despite LLVM's PTX 8.4 selector predicate.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F16 forms in the 10088-byte all-form cubin; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12220,12 +12220,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f16_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — LLVM canonicalized the four k32/F16 declarations without convergent. LLVM's selector advertises PTX 8.4, but CUDA 13.3 terminal evidence below requires the conservative PTX 8.7 override.
-    - backend codegen on `sm_89, ptx87`: `succeeded` — LLVM emitted all four k32/F16 forms in 2156-byte PTX with SHA-256 31144a1d5c93e7b11e5d74454366104c62db348b54e699715f2e41cedf2d680c.
-    - PTX assembly on `sm_89, ptx87`: `succeeded` — ptxas produced a 3624-byte cubin with SHA-256 3590a32ef95a1b4a7d4910cdf5542db6491c04eb9c76851ac8525ef3626a73b4. The aggregate fixture also assembled at newer sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx86`: `failed` — LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx87`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 789f05c3238b0f5a164d0da2615a031fbcf0ccefaf3ada07ad8316b9b465df55.
+    - backend codegen on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 727c802f25ad9448a068d9cb10d495d689000b8193301bbe622a7b5b367a4ba9.
+    - PTX assembly on `sm_89, ptx87`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 71bc48f0ece8bc2cabdc3510c94e6f76928640f0f114d403cfe142fedf82049b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM emits this form at PTX 8.6, but ptxas rejects it because FP16 accumulation requires PTX 8.7; diagnostic SHA-256 2244c15f79e1308d06fb3f86e84910358bf7569a16f1018b0d8a9983feed81da. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F16 forms at compute_89 and compute_90. Admission remains PTX 8.7 because CUDA ptxas rejects this accumulator at PTX 8.6 despite LLVM's PTX 8.4 selector predicate.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F16 forms in the 10088-byte all-form cubin; newer sm_90 also assembled. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12233,12 +12233,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f32_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — LLVM canonicalized all four k32/F32 declarations without convergent.
-    - backend codegen on `sm_89, ptx84`: `succeeded` — LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx83`: `failed` — ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx84`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx83`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx84`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F32 forms at compute_89 and compute_90; the admitted PTX ISA floor for this group is 8.4.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F32 forms in the 10088-byte all-form cubin; separate LLVM evidence proves the exact PTX 8.4 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12246,12 +12246,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f32_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — LLVM canonicalized all four k32/F32 declarations without convergent.
-    - backend codegen on `sm_89, ptx84`: `succeeded` — LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx83`: `failed` — ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx84`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx83`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx84`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F32 forms at compute_89 and compute_90; the admitted PTX ISA floor for this group is 8.4.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F32 forms in the 10088-byte all-form cubin; separate LLVM evidence proves the exact PTX 8.4 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12259,12 +12259,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f32_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — LLVM canonicalized all four k32/F32 declarations without convergent.
-    - backend codegen on `sm_89, ptx84`: `succeeded` — LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx83`: `failed` — ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx84`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx83`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx84`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F32 forms at compute_89 and compute_90; the admitted PTX ISA floor for this group is 8.4.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F32 forms in the 10088-byte all-form cubin; separate LLVM evidence proves the exact PTX 8.4 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12272,12 +12272,12 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_fp8_f32_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — LLVM canonicalized all four k32/F32 declarations without convergent.
-    - backend codegen on `sm_89, ptx84`: `succeeded` — LLVM emitted all four k32/F32 forms in 2761-byte PTX with SHA-256 6a663da2fe87fdd3a45698c9cd1748b39fe6b328f2fa81dd8d4f6a856bae4848.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — ptxas produced a 3880-byte cubin with SHA-256 289ad26a8680729e10aade2474fba52e60e3c0b845ee6ccb976d591a1f0d6642. The aggregate fixture also assembled at newer sm_90/PTX 8.7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_89, ptx83`: `failed` — ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
-    - PTX assembly on `sm_87, ptx84`: `failed` — ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 259330d34c78dcf13430648459b315462675a8b1fa9ec40bb9e06d84e171302f.
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d0c556e7b1590eaf7bd085ba902f4e397533f9512a18b2d9b57beef8a77e84.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 858a89c94a02717ace755a610d848cf949282519bbe039f74a0e9b5f904fd97a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_89, ptx83`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected k32/F32 below PTX 8.4; diagnostic SHA-256 34cd445a7598ba2464fab41cb0895499fd4d8857065f3d151a9f548654efb463. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+    - PTX assembly on `sm_87, ptx84`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected standard FP8 MMA below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — libNVVM emitted all four k32/F32 forms at compute_89 and compute_90; the admitted PTX ISA floor for this group is 8.4.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — ptxas retained all k32/F32 forms in the 10088-byte all-form cubin; separate LLVM evidence proves the exact PTX 8.4 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -12285,15 +12285,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_87, ptx93`: `failed` — ptxas rejected the compute_87 intermediate below sm_89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_87, ptx93`: `failed` — nvJitLink rejected the compute_87 LTOIR below sm_89. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
 - `mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12308,15 +12308,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12331,15 +12331,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12354,15 +12354,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12377,15 +12377,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12400,15 +12400,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12423,15 +12423,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12446,15 +12446,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12469,15 +12469,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12492,15 +12492,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12515,15 +12515,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12538,15 +12538,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12561,15 +12561,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12584,15 +12584,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12607,15 +12607,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12630,15 +12630,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12653,15 +12653,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12676,15 +12676,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12699,15 +12699,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12722,15 +12722,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12745,15 +12745,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12768,15 +12768,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12791,15 +12791,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12814,15 +12814,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12837,15 +12837,15 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 9cd4fd135bd7719c6a550526ba68c4015c8aefad2811e49005aefaa897359900 and 6dbf0669b5f2ce33bff37ff7e7c87879d755e61774637efb4770cf94591e0a16.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to ea1c5e21a22663af776dd799a3b07cb6dc4026faf0b8331fad61d87e8a061a11 and 395dcbc70858e8795c488013811f1b33b819851caabd05a15a0857fa0639a11, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 05a344118ae63c48bb38e531442bc1f8d4c78347d9a12c86fd2a949215b994ec and 46d809a8729b1140ecaebed9766db50f5b907e176634ff46b6dd8ff7ab1c122d.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 1a4b83a48257882b351d437b900e147628d84a8b5f1d57e051c6ece8e2d463cf and b36aee417d1cb416269e6310e650cd0e99db56b2e55c933a57af8e943d738044, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to 66c27c616ae715d3e1bb92faa78c88b84e450967ba167b7b8b30e7096c0762d6 and 93f34696c07351c51ab8ac186e5c0237f7ffb4017531b78f39d36fe01fed56d8.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to 4fa87b06b609066f59bcea622445bdde5cb6787d1f252825835c230f3962ab83 and 1fbd4a0fbf9d9143c5c7bbe1662b365dae10e3d8073ae286fe285eaa272ec946, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — cuda-intrinsics-gen candidate probe emitted all 25 TableGen-backed forms; the sorted manifest and PTX SHA-256 ledgers hash to e79a17cd8a0552d55eed1b2c17dc8a829d9990ead2fd4e55e7b8022dfc6a269f and b3362507676a2fdf271a42441fbd9f80abdaace5cfe567066e12f52ce9188f3c.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — ptxas compile-only accepted all 25 forms; the sorted CUBIN and nvdisasm SHA-256 ledgers hash to b643f185ca53f42090feb8cabe784356c7453a2bd008b9e677e6a5d4c6215219 and fda0e363142cd7421866859b220b03ce9165effd12bb5888ce14b699ce661ee7, with one matching QMMA type pair per form. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c50a4477a4848d8e1bd3753c09f64998baa687d07a673ac09a1cbac1f2b7872.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e199aedb32c46c0e6802b3bd963caa4963cd48278ed8c6d644cc9dd0466bb4c1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5743af0c719139e6b9748ddc9edad3a00057ccb6f50737daf495455826bf9d25.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cca9fd3d3bbc341c3658d109d37982dd600e2e514daf326e4133827f6e661ef9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79848de9174ee72596fe879eb4fa58562e252e82e64cad67c6e174e94ac9a02e.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1cf934d601fd4a574fdde418f45ba6b8cd0b1b0505ed6ebb6de1060ac40170d3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8cc3cacd357b8b98473fea3e778be515de2b6ed2c31a6384d26833bcb364df27.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fd8b4a41c30f73ebf06f471a904a538a34c220807bf6bf5d62046e5f04830b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; libNVVM release V13.3.73`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — libNVVM compiled all 25 cuda-intrinsics-gen candidate forms to LTOIR; the sorted LTOIR SHA-256 ledger hashes to 64961f12856a78fe21d451d9699ce57261d968d544e419ff2df4b554fe09abf5.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 701e011522fd8d4eeb3ee8a1d790c027da777c5d3625601bbafc485542a7331d, 6febe6b05a3c815745ca2bccf9e4c6e118e99cfff22325143c08c1acf6f4e3e5, and 0d03d94f65568b315fedd2d3dc89d9f0a424f851b97b9fd346a101edb9154ff6. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
@@ -12860,816 +12860,816 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — nvJitLink PTX followed by ptxas compile-only accepted all 25 forms; the sorted PTX, CUBIN, and nvdisasm SHA-256 ledgers hash to 071fd3ef0876391a20073d0e68ea24612619762b5821504bd3da19d8c9b56b1f, 8372c69f96306963b3566f25a0fb96318dc58a4c3f760e832c9d6fd2d1ef82ee, and af29dff1abb9124c325dca840b709080a3575da003ca93afe94233a8ff048c36. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — nvJitLink produced 25 kernel-rooted CUBINs; the sorted CUBIN SHA-256 ledger hashes to 15d2a6017e2457bab4a5831b3481c07e57bdaaca1383c13e64322fba384ff126 and every disassembly contains the matching QMMA type pair. Tool `libnvJitLink.so.13` reports `Cuda compilation tools, release 13.3, V13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c47aa6dd3e24383098f128a64c87c47ab4797f5e033e3fc4bb0a670d2503013a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c3aa0206bc409a773e92374c337ae29700fa54735792eeaafe1d8f87fcb9c523. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d9b3d3e50e66a2281bd4a40ade4118f0afb4d7789eef88afc79e9498ac5a3ea2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8ca1b55fd6cb49b5380574ff42ec7d55b6dbf557c08ff72873a5da7eecd90838. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab9e908618637e0d0699f72f1a3a80b33f56c41e57a5c802af03819ef2c5ef14.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7567be8747a7b101aa8df77028eef647926a3d0dea2f0fcd83d3ebe953a622e2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 81ae38ebf4bd8a59ff18a33203383a22cace76e383f473b31a6debf0f9ee947f.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 973c539e9cb7d01605f08ede31f1bcf9fb12155ae80857ea6da2f5bf085509f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact signed INT8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6524c9c1c5d31e51a7bc4f69ed88bb6f64bc9718c4cd4e6dbdf38d71cceb7406.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f67ea6f8fc76d2142c8297e2fbef29257d5a89eba91380d4ebbbf47cdce91489. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics example produced 11,096-byte LTOIR with SHA-256 71862b5aa53785a7f7ef9f7d86ddc5a4b88d893d6408ec2532f1f0a230173fda.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 15,464-byte cubin with SHA-256 955a6f99c580ad85d3c30a003f34f71ff9ccdd595386880f524daea4f1745eba; SASS contains IMMA.16832.S8.S8. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite s8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c512bacbd7f6da60ef9eda11b065823f600b9f0409a0ca562d8b2ae0e2c2df48.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57e571ed96c84be6aaa7c9428a16c4c002f791b55429f427820fcc3d6fbb177c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping s8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 77110080515c64834cb555469a0ea69ed12c499b30be821f8832fdcfac0f73c5.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7700bbce0406dc6977a0af2702a6679b1a41e88c867eed3903621b9d14beef53. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite s8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16e7d991475a0e8577e129c8337f9eb2ba08ec792a3500cd9890a5be58311b2b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 74351d962b7eb0ac0ca27e606b8d6e75468b39752b667b28e6277475de18bb29. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9cd0a0a84b94f37267419e33805b31860d5685fcd2cf4564783e28343a972748.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f2bac6e9f7ce9e6cb01846a046917c140500e7e876942100e7d3ce977cbfd90d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a51cb813a45b9cd0c0e9dc7038bec77a14194732f4b180ec96eecd3958d6a169.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3b2c092a2e9c78fd3580523dc6cc875e8903497e790eee28cfa15324c03685a7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ce1cd3a7682dc726d32cf180d1b39bdf9ac3aea8f12aa2a6c6423b7f521a15ab.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a5085e6a26e349f53eba46f4e630ecd2142786cbed096ed927acc94cc9c5d910. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 023ff2fe53a4cd8bb534791bde0e7384b5820f1fb7eb9cb1bed3627b8bed4c7e.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fedf24465346088c564e7971f6b7c29e92fceb052adb100fafb91b5e5b3b6bee. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping u8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67ac1fcd13f430b3a14dc6d8467cc9beec1a0b175c1e785fce2a9a91545d379d.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ba09c1ddf924a125f7171df66dec143a9c4d598837bdaed8de694fc7b52b4b8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact wrapping u8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ec734a83add935db593a1f90090b395a74af19c9c84661e61a80d46ba07966ca.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13d684b6400fcb97db851b1a163797e7b6cabd462f847f3f54b3342a0d49697c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite u8-by-s8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f2c069f3d47afb6653954d8f58fb146b9e927136d703104a92a2b8312c874d45.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fffbf71c999883fa695a7300a982460fc7e6425a322521cc9e98523dcfdc3fc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k32_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact satfinite u8-by-u8 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 eaabf735e8ebc1b813e38c3ec0d3c20f167786603c98ce8ebc1867abdf0e5d52.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 deefb4ba049125c5b91b39ac823909f05e24c773ceac2520b10210d11e02c17b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics module produced 22,492-byte LTOIR with SHA-256 54641033089cc8b470902881adb6e05dcf2dae80bf245fba1d3a313804a4be5d.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 17,256-byte cubin with SHA-256 eb2bea5cb70128e4b6f5dc07e5415ffd1500318732eed48d1f613f373cc5f04c; cuobjdump contained all 16 unique IMMA.168(16|32).{S8,U8}.{S8,U8}[.SAT] heads, including retained i0108. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k4_f32_tf32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - PTX assembly on `sm_75, ptx70`: `failed` — ptxas rejected the form below sm_80; the diagnostic SHA-256 is 73ef62d63bacc273dbfc152d2cb3acb93bfe3c744b3003e49cf66cd8a47e0b5d.
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - PTX assembly on `sm_75, ptx70`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. ptxas rejected the form below sm_80; the diagnostic SHA-256 is 73ef62d63bacc273dbfc152d2cb3acb93bfe3c744b3003e49cf66cd8a47e0b5d.
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5775-byte typed-pointer IR fixture with SHA-256 42a32e7fc0a6d40513480eff6d1fef5603bd2dc07621b1a1e3d8d946dac1f8e6. It emitted all five forms once in 6449-byte PTX with SHA-256 b7d87c4199fa520ab749fe1dd8f3b37a3b6685c9a7a9e1c8cebbf99342d2bd50 and 3676-byte LTOIR with SHA-256 582cd60f2d911bc38214437d619a334cb8e5cb96d6ec811329866aea4d027855.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced an 11048-byte cubin with SHA-256 e89808e3781dfa9b928f239cf2943777417f6ad789fb6c10f6981777aff6d4ff. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced an 11048-byte cubin with SHA-256 1d242578ce7378f849df0dbcd28bd43a60af64b7f76359b7e5510ecd23a14572. Its 52dc9bf332bdb595b2910550d95f0d0f8c44477440c52977bcc6b0699cecffcb SASS dump contains exactly HMMA.1684.F32.TF32, HMMA.1688.F16, HMMA.1688.F32.BF16, HMMA.1688.F32, and HMMA.16816.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5ff5fcfbf7af82721582a788d144fe6745bd1ea783aef2ff9102a328612ce3e.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 95520ee8bd9137e246a8f6b10d90bf414fd6640b6941209df067082273bab6a4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5b42070729512dc4a34d959933cb2ed579c549ca2c4a34ba53634c793c2104c5.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 052fb66d0981ed2361e6cde56d62007088995b7ba713d6da1691e0012e7f4923. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a4b6f9f1214b02bafdfe07cca12072fdbfab865362fd3b295e688a738be8d0df.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 814a279d3913788ca9cc088325b84198c97e9693aaa13778c28f0f654b834e38. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6d117f7dc3771deff164ddacd83b64a80d1b9c3f33543c2323394f860296b3a.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 232ee3454e7542b49e296aafcd4c9f38235fe2141fb224ca3f329aec36d456b3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 48af8df444ba8fcff6c181e3862bbcfabfeb9817b970ddb66ea95df74814252e.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7e02b52b0a757ec780b5e93f59ff1abdea9770a82d4f0f4d1384035db9f803b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c9140abf7847a4b891673bba722aaef6a63ce9520ae3764efd411de6d636208f.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3824b5037253381a849ded485ed000a3c459f68a94dcd43b809a72958bf618aa. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa313e124cffc714546231b65e670c63303446e7edb95dce4af94ca57bf258a3.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ed63a17572d63f68c5f40f43cf5920d3432723ec6b946bd1689e3c44d3fe4f9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k64_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The pinned LLVM 22 probe emitted the exact instruction in C, A, B API order and retained convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX at the exact sm_80 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 59332065be960c742528cb03bad200c2c5a5ad472285420dd46116d3d9da8d88.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2b6ed03aae00d63ec51d849f1382eb95b8580686067589bea8eaa9433e3d8be1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The isolated floor probe produced 34,566-byte compute_80 NVVM IR with SHA-256 c30073aa40593550a04d30837b570e0fcf40dd71136e9edccd48e33dcf6c4ac9 and 10,800-byte LTOIR with SHA-256 166e339fa1cca7d02147c6fab6d3c8b36ffcc38b2b4848c7ac024f44470ea978.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,880-byte cubin with SHA-256 815f33adccc2e248534b3c69d650d9e2ca5c483846eb8ea6663d909cb03eb785; cuobjdump contained all 16 distinct IMMA.16832/16864 signedness and saturation heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k8_f16_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5775-byte typed-pointer IR fixture with SHA-256 42a32e7fc0a6d40513480eff6d1fef5603bd2dc07621b1a1e3d8d946dac1f8e6. It emitted all five forms once in 6449-byte PTX with SHA-256 b7d87c4199fa520ab749fe1dd8f3b37a3b6685c9a7a9e1c8cebbf99342d2bd50 and 3676-byte LTOIR with SHA-256 582cd60f2d911bc38214437d619a334cb8e5cb96d6ec811329866aea4d027855.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced an 11048-byte cubin with SHA-256 e89808e3781dfa9b928f239cf2943777417f6ad789fb6c10f6981777aff6d4ff. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced an 11048-byte cubin with SHA-256 1d242578ce7378f849df0dbcd28bd43a60af64b7f76359b7e5510ecd23a14572. Its 52dc9bf332bdb595b2910550d95f0d0f8c44477440c52977bcc6b0699cecffcb SASS dump contains exactly HMMA.1684.F32.TF32, HMMA.1688.F16, HMMA.1688.F32.BF16, HMMA.1688.F32, and HMMA.16816.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k8_f32_bf16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - PTX assembly on `sm_75, ptx70`: `failed` — LLVM selects this form at its imported sm_75/PTX 6.5 predicate, but ptxas rejects it even at PTX 7.0 because this BF16 m16n8k8 form requires sm_80. The PTX 7.0 diagnostic SHA-256 is 69f7da673aef781e3eb0c06d602ccab1fdc5d95e4b0d645d6e118fd55d0747c8.
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - PTX assembly on `sm_75, ptx70`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. LLVM selects this form at its imported sm_75/PTX 6.5 predicate, but ptxas rejects it even at PTX 7.0 because this BF16 m16n8k8 form requires sm_80. The PTX 7.0 diagnostic SHA-256 is 69f7da673aef781e3eb0c06d602ccab1fdc5d95e4b0d645d6e118fd55d0747c8.
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5775-byte typed-pointer IR fixture with SHA-256 42a32e7fc0a6d40513480eff6d1fef5603bd2dc07621b1a1e3d8d946dac1f8e6. It emitted all five forms once in 6449-byte PTX with SHA-256 b7d87c4199fa520ab749fe1dd8f3b37a3b6685c9a7a9e1c8cebbf99342d2bd50 and 3676-byte LTOIR with SHA-256 582cd60f2d911bc38214437d619a334cb8e5cb96d6ec811329866aea4d027855.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced an 11048-byte cubin with SHA-256 e89808e3781dfa9b928f239cf2943777417f6ad789fb6c10f6981777aff6d4ff. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced an 11048-byte cubin with SHA-256 1d242578ce7378f849df0dbcd28bd43a60af64b7f76359b7e5510ecd23a14572. Its 52dc9bf332bdb595b2910550d95f0d0f8c44477440c52977bcc6b0699cecffcb SASS dump contains exactly HMMA.1684.F32.TF32, HMMA.1688.F16, HMMA.1688.F32.BF16, HMMA.1688.F32, and HMMA.16816.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k8_f32_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — LLVM canonicalized the five declarations to nocallback nounwind memory(none) without convergent. The 6254-byte source and 6413-byte canonical IR have SHA-256 883f351fb1d6eebd787e1312c2365c2dd5ff5a4ad852a6a8ee71901b3dc2ee16 and 20c9a527d322dea1f871debfa500120cccdb8ab1174fa73403f6bfc0f57a1992. O2 merged ten typed calls to five, while ten convergent side-effecting inline calls remained ten; the typed source/output hashes are 139d3203e4622002bf662064a1e83e7c1b4a8c57a82854bfd24fb7a7831d2e88 and cd57bd9a4f2204f7bd69b71fde7e709e183d7815e9e708c786d9c3dd7e087ac8, and the inline hashes are 21548f313f29c146ea089c082ccf5903ac17777dddd7bd663dbd9b59ea53720c and 040d2cc2dc6029aa767212a4b44cfd430c557ee81c2e76bb5ffa740a0a3cd691.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned LLVM emitted each exact MMA form once in 7366-byte PTX with SHA-256 4da1521d328388d9b779f0f070db4d51fe65127288f62a1d69c09834d444f610. The typed comparison fixture also emitted all five forms, but is not selected because it lacks convergence.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted all five forms and produced an 11048-byte cubin with SHA-256 c09d14bf15e6954e1d506e002f53eba2aa3e770bf501ba43346d28587203dc29. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7408de871f04f7b802723edce4d9ff6b805f77705eaa3167cd07096db1af3a.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 680047c80107706d8c0efbc9bdae4b2bbf1b691bd8448c7fe94ae5590a342abe.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 5 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 979c823fa344ced3e6a13cb42e0fe1380d5f120da81df94e11913a1601c360e4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 5775-byte typed-pointer IR fixture with SHA-256 42a32e7fc0a6d40513480eff6d1fef5603bd2dc07621b1a1e3d8d946dac1f8e6. It emitted all five forms once in 6449-byte PTX with SHA-256 b7d87c4199fa520ab749fe1dd8f3b37a3b6685c9a7a9e1c8cebbf99342d2bd50 and 3676-byte LTOIR with SHA-256 582cd60f2d911bc38214437d619a334cb8e5cb96d6ec811329866aea4d027855.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas produced an 11048-byte cubin with SHA-256 e89808e3781dfa9b928f239cf2943777417f6ad789fb6c10f6981777aff6d4ff. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced an 11048-byte cubin with SHA-256 1d242578ce7378f849df0dbcd28bd43a60af64b7f76359b7e5510ecd23a14572. Its 52dc9bf332bdb595b2910550d95f0d0f8c44477440c52977bcc6b0699cecffcb SASS dump contains exactly HMMA.1684.F32.TF32, HMMA.1688.F16, HMMA.1688.F32.BF16, HMMA.1688.F32, and HMMA.16816.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m16n8k8_f32_tf32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact TF32 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 229d640dcb7295b409e4d690f6094a704af0b77e69e834abb5c2c9d26ac5daa0.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7543fed70ec348791089d56f1eee6d51e13501a1a40092f118bec2c99b734d3c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics example produced 11,096-byte LTOIR with SHA-256 71862b5aa53785a7f7ef9f7d86ddc5a4b88d893d6408ec2532f1f0a230173fda.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 15,464-byte cubin with SHA-256 955a6f99c580ad85d3c30a003f34f71ff9ccdd595386880f524daea4f1745eba; SASS contains HMMA.1688.F32.TF32. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k128_s32_b1_and_popc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.1 and sm_80 floors. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6cf34f19543565ab3b0a268c1b38020771261bc358df589709391ec5e99ad3f.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93107316bf0cf5ef84b40e52ec5ef683b2781a3e3a0e75737e4892184d0c6457. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 165,143-byte compute_80 NVVM IR with SHA-256 50e40ba550994145a6c57741b22786e09252da4d786c0e296d8de14cfe44d30e and 47,744-byte LTOIR with SHA-256 060d4347bb268f38e14ebe8c5ad9db64fbf1781d578f14c8799f3d8f54f2db71.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 20,456-byte cubin with SHA-256 aa2ce30ecae34abe738c760e7d3f2b20d35744652afdae0b6abb29b4bf10d8a8; cuobjdump contains all six distinct BMMA shape and AND/XOR heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k128_s32_b1_xor_popc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx70`: `succeeded` — The pinned LLVM 22 probe retained the exact instruction in C, A, B API order as convergent side-effecting inline assembly.
-    - PTX assembly on `sm_75, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the exact instruction at the PTX 7.0 and sm_75 floors. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 666182ec743c9ffe0422975ab4ab4faf2e3ddcff656913f34303f15200fa3700.
+    - PTX assembly on `sm_75, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 452782b4303bcdb3357c76efeae34f913d781f5673cd5a97eace443e1dc28b77. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated exact-floor route produced 2,613-byte compute_75 NVVM IR with SHA-256 c834c251fc1e95e0eaec0f8e60a572e9daa5c8e3d6ce80ff7eaf6d2757786871 and 2,336-byte LTOIR with SHA-256 90f5d4bd0fa9d90e8226eb99a588cb5d65d6eff6d07443d59905cda845f2a846.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,112-byte cubin with SHA-256 85c259e08b117762519d418f29d9756e089f123aafa163bda467c62504049d51; cuobjdump contains BMMA.88128.XOR.POPC. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact wrapping s8-by-s8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5c2da0b56637465b6318014f3bb75f68d843c14ef827e70a7ef937625ad5e2c.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 df22078cac5b20809b7d5a9eea6a29d6cfe354500d77fa360d86a0b4a1023852. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact satfinite s8-by-s8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 270630238dd5c5f9cf517a25b950096e73e12e4479f81ac8c1cd3283017390d0.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1bf5493b1e849d63814126e394e6f0572529d702a2433d0de1ff47b7b4ac0650. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact wrapping s8-by-u8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab1d12b93c4b140169d3169350a16500008aa4163f09614a631b7e1dc4fdbcb1.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a423b06a09e13590b1d54c6f8d8be51d1e394145354b004a635293e1d76ac83c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact satfinite s8-by-u8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d10485a6ecf25b1ef218ba5ca3641e58f8066afb750fe46d83566b3fe2b4f072.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2f5aaa3aac438dbac75e659a7a874e86e99e3f0114d38a53f3d889f2063ea0b6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact wrapping u8-by-u8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f6d577d18158476e7ab49d17a2ad0f47da8b6ae752a6ed080d20e1380d52bee.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1955df76beb6ee81ed7ea879c3200f4f6bb7e1966deb92e945b3f325841dc5cd. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact wrapping u8-by-s8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6f75f6988bc1dda288d7c571d72462c4298a17b37aaa830a033de583852ac772.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4dff1215835b5b26444fd0e25f94789efea21a227da12d555c889c796c0ea93f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact satfinite u8-by-s8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 93bd9995092da3436c05eb1720405ceeea64728cf2eb687f907c7b1c820d0148.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 67dbf19d3684c67bd2b2ffface224d08f9e24a216e617ebf5307e530bc492d97. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k16_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The generated probe emitted the exact satfinite u8-by-u8 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 51db7a6fe697ef26e445bac4abdaa4d5e45e887e8aa5a366f608654d3466aab0.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06aafc80a8ca8190b809819eaac9c4776807f6380a05ed5a02ce479898506506. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,109-byte compute_75 NVVM IR with SHA-256 8658db3fa9259db3b16856d2012d413ad6efcf2120c5df7e6fbcc0a0801b4406 and 4,948-byte LTOIR with SHA-256 eb153d1b19d2c770389d5e9ab503f524f3bd220052e56980b3a3134f356b6613.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 f731151e06c4c452de06d36d4aeb627e6f8e2597ba015a8fc53f58f4956dee58; cuobjdump contained all eight distinct IMMA.8816.{S8,U8}.{S8,U8}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping s4-by-s4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 242a7d2104a977a3c344c947eb5c03d10a18711cb776b10a27c2519d3760eeb0.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c5b6ee75c034c9af33c18ab6c91f725b42d2b54c8bf132ec1b89c115cb9a3548. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite s4-by-s4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b5fb1110c752bf4fe08f37bd8b7f8345d676ca3eea504e71bed250fb3c7497ce.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5e9c1003612b5e1b0441919a9d51eeadbd9a9c26f365be978cf5d279b0b0d1fc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping s4-by-u4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 57e936cd723c8522e8d7ad7c20eec7f1efe473d9dde5381bc74874bbd737acc1.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 540c2a2b1c2c2945022dad6f87d3cdf948ade2eab7a1aa2149e47b320adf4e48. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite s4-by-u4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1a0ad568976eb0b009eeb8941f96e74169d76cd8f5e486c761848b16a6a78f68.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ee342c6e9f3f578c32c6af5dd1c2668cf556669258abfcce6f85298f77a2fe90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping u4-by-u4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16ea31bd434789327320d0683d1cc83a11058292365655503c77c2143139e1e4.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75676b7cb4e4fb123537b9f8f1cc40ccb04e1db0271ab3eb760961f4ea73ea37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact wrapping u4-by-s4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a777e8a890e81f7bc799e08ffa31129ff28d46f13cd9bbb9467f7d3e09e294b6.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 87501053789841d181fd1aae8dac5b4dc516639d99983f1867fd6f4df9be4836. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite u4-by-s4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d3a2ccb17359a5a84b5f2ffbbbf21bdbf3e862d6a2a6e007cba7442aa65144a9.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 787d75bc93a0a84ac830fa9dd7472adc96e9b78362d7e6ce1b107301d5c6d0a6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k32_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx65`: `succeeded` — The LLVM declaration lacks IntrConvergent; the generated probe preserved convergence and emitted the exact satfinite u4-by-u4 instruction in C, A, B API order.
-    - PTX assembly on `sm_75, ptx65`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.5 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9b77eb2760494d673de880beb0e9da2110a690897b8968ba0b33d57e0f7b635f.
+    - PTX assembly on `sm_75, ptx65`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 666fe69ffd45aab6a3f5bfc5183c41460a862957228ada97aee5dece4d4b92a4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.5 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — The isolated m8 floor probe produced 11,127-byte compute_75 NVVM IR with SHA-256 2e7b1ba07e4a33d1e8409480ea887f824bc1e4a4a7e5d9d2eae9d96302bf5278 and 4,956-byte LTOIR with SHA-256 a5984684763bcdda15f246065a3f8de5d38e803558bd5a7303322343e1a15b33.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,240-byte cubin with SHA-256 ed0d8aab36380a3c8d47906de00491623462d5b4ac9ee3c77e6fc46b17c18963; cuobjdump contained all eight distinct IMMA.8832.{S4,U4}.{S4,U4}[.SAT] heads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_m8n8k4_f64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The generated probe emitted the exact f64 MMA instruction with C, A, B operands kept in public API order.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX for sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 455ca35747a458536fae20c94d1749e6b7c07dc9b482fb469192376c9e10a1a8.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f7bf4efce75e805867c6ef541acdd239b2df1f6819376e1d08b24cf76398b507. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The generated_intrinsics example produced 11,096-byte LTOIR with SHA-256 71862b5aa53785a7f7ef9f7d86ddc5a4b88d893d6408ec2532f1f0a230173fda.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 15,464-byte cubin with SHA-256 955a6f99c580ad85d3c30a003f34f71ff9ccdd595386880f524daea4f1745eba; SASS contains DMMA.884. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f178dc6498dfe1335f303815c77a74f6c64eef9102c2ae74312466d8ac126ab0.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e32824932666209d5959192f45f412c7f087fd40cd3b8a97e82e9004721855c5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f05e41f54c80df9399f1535e4630d9ac644c42c0c8bb66c19902effd9fe96f3b.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2733247c94953620e2bdfdda7e22bb9f6a0f731285caa41a50d70a50d63c92f3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 751f4ccde1a953061a6de9f03fc03e29bcab215d3e0dfaf5cadec2fdab2ffda9.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4b7f6a5ffd37de7a86d2c35f122c69356b19e3ab05c2eaf64e726dd677fffd2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e91aa50a469e4ae22cabf614f0c86c11a381dd99b7d57a4f376c7f86b815819.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c0e9109eafe381fc83456148a9518994753df9868388321037f0cbe11c64a260. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 aa743b0396151d7184adfa37f3ff722cd693b7bdb74f433a522498874947fbfb.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 56253d8e8cd56f03dadc8369add61af230eac577a3258c42a28dcf4f2ede4f3d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6c8d0ebc3993fe3649feb8dc574ca4aa6d9bbb8a54b57bab64439c72b3183f91.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e93fae493c1bff6c15d4b462fe8d3e9926f99da8c339f07247d5633515195e36. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75d63c6f58dbf64b7ff7c0bff1b89fd0c2c7776ecc86997ccfcd3184130f610f.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3f4ba91d089260787eb84c1e93a4b9ad6ea30660ea314e755f025574e205657. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k128_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 4,867-byte combined public-order inline-PTX fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) covered all eight m16n8k128 variants with selector 0. Pinned LLVM emitted 3,913-byte PTX (SHA-256 0f39eba80daa327cf5276f2942e940c5f2df3cab586db75af10210b21fffdfd6) containing exactly eight standard m16n8k128 instructions. An auxiliary 4,528-byte typed LLVM fixture (SHA-256 3d5971c2a14df66be17427c8da533c2f29745173ff867b366eeac739495959c5) used A4, B4, C4, metadata, selector source order and emitted 13,602-byte PTX (SHA-256 601e40fc487232c248466c5a0cc688cb85cee2cac45e2a571991438dfa78aa23) with the same eight forms. The imported record has fourteen i32 arguments, four i32 results, ImmArg<arg13>, IntrNoCallback, IntrNoMem, and Range<arg13,0,1>. Pinned LLVM rejected selector 1 (580-byte fixture SHA-256 a810dcb7589210cd2ea8d46d261c0841dc4207f984c283bb4f7f7ebe29c68665) and a dynamic selector (603-byte fixture SHA-256 5573f9ea9fd0f07669c5f54af9501332933b66a4ff7bfbcc103f0f4cf2b498d4). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selector 0 at PTX 7.1/sm_80 (258-byte fixture SHA-256 308e16eaf954fa79243181ba7e91c28f5310e1a5b275e80092bd87458c2d48b8), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 50f8efe377dee2b7502120e012557548b6af95f53fcb395ae1d34060d0169b9a) retaining exactly eight IMMA.SP.168128 instructions. It rejected selector 1 (SHA-256 7bfbafd9e5a67d3de2bd4e0bf55b4be6865569c374bdc64962e184385b353679), a register selector (SHA-256 93909da87dbe1a1b12a7064b4b0b4103c710bec837ab605326091e8888b10bb5), PTX 7.0 (SHA-256 b3a44520c93f1500442c99d3eeaae69e826fd28c3afd6f7f9345d3ce2a26d37f), and sm_75 (SHA-256 46c14a11d7814b7a83f082246218f5c8c5107844cd494958191cb9bd0019d6e9). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f13fef29cc4b67af7e5a181bef0b806732d7eea0ec0f5176e91581b982f20511.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c0557fe188cb6b3e8cf492411a8e5ae3072d44fd5ab349e21951c62b0ed30a9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 4,867-byte NVVM IR fixture (SHA-256 98ef7b8e03f9f690c118f360514f74e6da85151fd83b1be71c8b102083abb76e) passed CUDA 13.3 libNVVM and produced 3,811-byte NUL-terminated PTX (SHA-256 595ec2f639ccb02e83580254efa78758160ccb8148198a08b932b9f7e2962afa) containing all eight variants with selector 0. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 3,810-byte PTX (SHA-256 302d6da269cd36de7991678afcc1b0c13be08302e4338bf9a170831d0e7f4dea), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 dca42ca41d8127353ab0a9a2aa1673bc9d5a1deff86d0c0aed18b7b5c88c4572) with exactly eight IMMA.SP.168128 instructions. The PTX 7.0, sm_75, selector 1, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 e304b99f1ddaddb561a7cf45c39fa3b97c5252091ffafbcca7577dfb60d395fc retaining exactly eight IMMA.SP.168128 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping s8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6117422b739d9c39ae85c36a064b6ba0cfe6242cbd6c19fffe7ff68365e00a79.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a84718234f047484ec8d86c72e2a333a2982ea25b4266f109070266baa49c715. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact saturating s8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3c5d9bb6256f7009234918046b2dfb11507265c220722e1a42aa20653a409aec.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9c0a5d19a7e1637ff335da9a214e99e06463d5160df778e1bad3dfcfc6874890. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping s8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 32f1c01ee1a43ca81812d076ee8db48b8928aa441daf3f71dfc025f6b50a516a.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 875de648ec423e0f311608024d48bd81ae45ae442c9741a8f08c01502f863e9b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact saturating s8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b79bb4f92abfb0ce7a3b13baed56134e8043e73d4552d305d42f8f2d97661bfd.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 adbd36486601138ffe021f493473ae535351c523380e780a11678e1a4da0e4e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping u8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34822f024cc535508a1d7ae0c22eb25c61a337e07e674e4c5d0821c85d2e44e0.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c0cad4768d71256792c04fdf7e962f3630275a62a909d6ecd15a67a1aacb7765. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact wrapping u8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1dc739b2f8f0bc7f5d2a744a638f9f9e93a9c9763584bdc4ef239e3b740a7900.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11793a16a1edc4a91afc40987b925e8829ac10289034931cf23a5dd19c54ff17. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact saturating u8-by-s8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d51f3473017b1bc3eafcae85544debb79cd50a621baebf76491730e6af4dfd83.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c8b4cd0226a362e5b29976f9f58f7fba85de7a0f25858de6e77bef814660815e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k32_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The pinned LLVM 22 combined probe retained both selector forms and the exact saturating u8-by-u8 instruction in C, A, B, metadata, selector API order. Its 13,238-byte IR had SHA-256 5652a669e731467a3c950b9b1aecc39e51156df2f6329a112c52569e97cd5a8b; the 39,599-byte PTX had SHA-256 4138b4c3d39aabbf987b868206ef03b4b135fb5999c31420c5c6780d11d3ab4f.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instruction and selector combinations at PTX 7.1 and sm_80; it only advised ordered metadata for future performance. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d8c7f3a382ee9d1ea29abdb323d012394df4bf6bce1fa24e8594c78e8bfa65f7.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46586ec2c81c1e15a80b74485e9e14cad94c67af27deead4a2ce433787a814af. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 189,884-byte compute_80 NVVM IR with SHA-256 c09e93484e59f6e1b76a62db146f66164234f6f2fb73842cbad12a14d6cac76c and 54,308-byte LTOIR with SHA-256 d8b822ccfe9e6918437f8fdc6b2bcfe5c1326100cfb135306d1d4df4dd5b01e6.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 21,480-byte cubin with SHA-256 ba48413405086ccf4aa28fba90fb7e40c0dc6df25af437a981ef80e89a44d3bd; cuobjdump contained all eight raw sparse variants plus both representative compatibility-path calls as IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_f32_e4m3_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_89, ptx84`: `succeeded` — Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.73`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled a 4-function NVVM IR module (SHA-256 42a84360a8efa402dae317b5fcb41c95cd857eac174eadf3fc01458b1c513504, requiring the !nvvmir.version 2.0 metadata and an explicit NVPTX data layout) with -arch=compute_89 to 7,193 bytes of PTX (SHA-256 28b1418b52d7911cc7122424e57c4fa0814b51bde354a69d7fa9f1dad3693dac, .version 9.3, .target sm_89) retaining all four requested instructions verbatim, and with -gen-lto to 3,092 bytes of LTOIR (SHA-256 846592e970ecc3f8a8ede212a9c4b5f87586e3629c0ee2b151dab6f62114ab1f). The libNVVM route for compute_89 is accepted: no capability error was raised for any of the four forms.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the libNVVM PTX at sm_89 into a 1,464-byte cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a), byte-identical to the cubin the pinned-LLVM route produced from the same four forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — nvJitLink produced a 1,464-byte cubin (SHA-256 18651b809387f94d6074dbdb66590859fac9169a6aab7650569ebd287e8e53b1) from the 3,092-byte LTOIR. The device-link step required the explicit -lto option; without it nvJitLink rejects LTO-IR input with rc=18 "LTO-IR input cannot be used without LTO enabled (-lto)". Tool `/home/kxqandccx/pr-e2e-20260918/tk/lib/python3.10/site-packages/nvidia/cu13/lib/libnvJitLink.so.13` reports `Cuda compilation tools, release 13.4, V13.4.92` (SHA-256 `3afdb9677477dd584d3712a9bf2978d085729f0ed6e6432027e925d1e312e3c9`). Artifact `cubin`.
 - `mma_sp_m16n8k64_f32_e4m3_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_89, ptx84`: `succeeded` — Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.73`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled a 4-function NVVM IR module (SHA-256 42a84360a8efa402dae317b5fcb41c95cd857eac174eadf3fc01458b1c513504, requiring the !nvvmir.version 2.0 metadata and an explicit NVPTX data layout) with -arch=compute_89 to 7,193 bytes of PTX (SHA-256 28b1418b52d7911cc7122424e57c4fa0814b51bde354a69d7fa9f1dad3693dac, .version 9.3, .target sm_89) retaining all four requested instructions verbatim, and with -gen-lto to 3,092 bytes of LTOIR (SHA-256 846592e970ecc3f8a8ede212a9c4b5f87586e3629c0ee2b151dab6f62114ab1f). The libNVVM route for compute_89 is accepted: no capability error was raised for any of the four forms.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the libNVVM PTX at sm_89 into a 1,464-byte cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a), byte-identical to the cubin the pinned-LLVM route produced from the same four forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — nvJitLink produced a 1,464-byte cubin (SHA-256 18651b809387f94d6074dbdb66590859fac9169a6aab7650569ebd287e8e53b1) from the 3,092-byte LTOIR. The device-link step required the explicit -lto option; without it nvJitLink rejects LTO-IR input with rc=18 "LTO-IR input cannot be used without LTO enabled (-lto)". Tool `/home/kxqandccx/pr-e2e-20260918/tk/lib/python3.10/site-packages/nvidia/cu13/lib/libnvJitLink.so.13` reports `Cuda compilation tools, release 13.4, V13.4.92` (SHA-256 `3afdb9677477dd584d3712a9bf2978d085729f0ed6e6432027e925d1e312e3c9`). Artifact `cubin`.
 - `mma_sp_m16n8k64_f32_e5m2_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_89, ptx84`: `succeeded` — Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.73`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled a 4-function NVVM IR module (SHA-256 42a84360a8efa402dae317b5fcb41c95cd857eac174eadf3fc01458b1c513504, requiring the !nvvmir.version 2.0 metadata and an explicit NVPTX data layout) with -arch=compute_89 to 7,193 bytes of PTX (SHA-256 28b1418b52d7911cc7122424e57c4fa0814b51bde354a69d7fa9f1dad3693dac, .version 9.3, .target sm_89) retaining all four requested instructions verbatim, and with -gen-lto to 3,092 bytes of LTOIR (SHA-256 846592e970ecc3f8a8ede212a9c4b5f87586e3629c0ee2b151dab6f62114ab1f). The libNVVM route for compute_89 is accepted: no capability error was raised for any of the four forms.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the libNVVM PTX at sm_89 into a 1,464-byte cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a), byte-identical to the cubin the pinned-LLVM route produced from the same four forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — nvJitLink produced a 1,464-byte cubin (SHA-256 18651b809387f94d6074dbdb66590859fac9169a6aab7650569ebd287e8e53b1) from the 3,092-byte LTOIR. The device-link step required the explicit -lto option; without it nvJitLink rejects LTO-IR input with rc=18 "LTO-IR input cannot be used without LTO enabled (-lto)". Tool `/home/kxqandccx/pr-e2e-20260918/tk/lib/python3.10/site-packages/nvidia/cu13/lib/libnvJitLink.so.13` reports `Cuda compilation tools, release 13.4, V13.4.92` (SHA-256 `3afdb9677477dd584d3712a9bf2978d085729f0ed6e6432027e925d1e312e3c9`). Artifact `cubin`.
 - `mma_sp_m16n8k64_f32_e5m2_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_89, ptx84`: `succeeded` — Pinned LLVM lowered a per-intrinsic probe matrix at sm_89/+ptx84. Each module is one convergent inline-asm define; the four sources have SHA-256 0e19213f66bf48490191dcb26cec4fbcc9b3811b9d470d720c7d3ca6dd75fe92, d1488ca123f1d551a8d11054e757bcb996c9b46288387472a5c1deec77ad6d3e, 7fc0a1505fdbff6b417c104d0e86630c5b7646cacc26de1362829305c520f8b7, cc638d7394b354e17d0e2e226ddfd9329b70eba4da6f4e5d757f41b8a1e8ac8d and the four 2,905-byte PTX outputs have SHA-256 422fc5b0a05111aa642045eb5457f4c2015e333680893446e36a1ce5219cce76, 75fcba9ddcd36eadc1f2fda9545b99b722b7176b741415f8e6e8308885f314ab, 277ab86e14079870065530ce63f8ca013ce7010ae586d345de5e1d178c3a8560 and fff31942daac87bc05a1113d194a660980ace3ea0f03b00f89142faad53810ba. Every module retains exactly one unguarded instruction of the requested form with 4+4+4+4 register operands, a metadata register and the selector immediate 0.
-    - PTX assembly on `sm_89, ptx84`: `succeeded` — CUDA 13.3 ptxas accepted all four per-intrinsic probe PTX files at PTX 8.4 and sm_89, and assembled an all-four-forms fixture wrapped in a .visible .entry into a 3,368-byte cubin (SHA-256 6055d90ad156664d015627200c40b8d77cb9853ed3d87dce9fcda2a1461e4e78) whose .text.sparse_fp8_all_probe section is 384 bytes of real SASS. The per-intrinsic probe modules carry no .entry, so ptxas eliminates their bodies and emits a 1,464-byte stub cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a); the already-admitted dense FP8 MMA family produces the identical 1,464-byte stub, so the stubbing is a property of the single-.func probe harness and not of these forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.4 / sm_89+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b4291dcecc12bc81abeb0be6aeaced6d49643bef41540dbf09a3351dd88cd047.
+    - PTX assembly on `sm_89, ptx84`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9112d0e3f69969f98e685af0f6770243c98fd74ebfe159b84dbccb33abff93b4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73` at PTX 8.4 / sm_89+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.73`, SHA-256 `e55af1135e987372355d19d66cb74e1a1e99c32ac98ffad14fb567eef16e4f77`)
     - backend codegen on `compute_89, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled a 4-function NVVM IR module (SHA-256 42a84360a8efa402dae317b5fcb41c95cd857eac174eadf3fc01458b1c513504, requiring the !nvvmir.version 2.0 metadata and an explicit NVPTX data layout) with -arch=compute_89 to 7,193 bytes of PTX (SHA-256 28b1418b52d7911cc7122424e57c4fa0814b51bde354a69d7fa9f1dad3693dac, .version 9.3, .target sm_89) retaining all four requested instructions verbatim, and with -gen-lto to 3,092 bytes of LTOIR (SHA-256 846592e970ecc3f8a8ede212a9c4b5f87586e3629c0ee2b151dab6f62114ab1f). The libNVVM route for compute_89 is accepted: no capability error was raised for any of the four forms.
     - PTX assembly on `sm_89, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the libNVVM PTX at sm_89 into a 1,464-byte cubin (SHA-256 b1f806ccaa8a73aa80b76862b643265a508494c53028348b0f4e730d7a86847a), byte-identical to the cubin the pinned-LLVM route produced from the same four forms. Tool `/home/kxqandccx/pr-e2e-20260918/cuda-13.3/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
     - device link on `sm_89, ptx93`: `succeeded` — nvJitLink produced a 1,464-byte cubin (SHA-256 18651b809387f94d6074dbdb66590859fac9169a6aab7650569ebd287e8e53b1) from the 3,092-byte LTOIR. The device-link step required the explicit -lto option; without it nvJitLink rejects LTO-IR input with rc=18 "LTO-IR input cannot be used without LTO enabled (-lto)". Tool `/home/kxqandccx/pr-e2e-20260918/tk/lib/python3.10/site-packages/nvidia/cu13/lib/libnvJitLink.so.13` reports `Cuda compilation tools, release 13.4, V13.4.92` (SHA-256 `3afdb9677477dd584d3712a9bf2978d085729f0ed6e6432027e925d1e312e3c9`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ae995db003010942ebdf8a541fb71e7a209b14d2035dc9b39ceebb44a20a8b22.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d3e0072eed343ab18506217d990ecc273e425433b2c7866bbf1cb84c84c48651. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0d1ce7e2d45fa78ac932971fa7ccc0f46e00d454de87b3e1ea926b0b22636b9b.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f8e6cb0ffef1b7c51d3caa4e5f589992e341d54ef3c6b0f0d1e8ac2581ae5db2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 512623d519f30b4de0fca57417959c5a3e4f6208ac12e3691744f3fbec21cefa.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0e549ea726768ab22519dacc89c24c3546e6cfcf709f8b53b5a8e86030aa8deb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping s4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 402335a93decd5671794a0c5ec5e6f09ffd69ed6b26754fac45e7f294b3aac25.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 11e848685fa3b253ee0b3abb5fee26cf6fb7b3e7f567351483da94bd54ed1afc. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating s4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping s8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fcd1b6080f311a524c25afacc87c7db3e52597ff2da06494ecc63a319740ff3a.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e7c617c1291bbecb8b6aa72f56558797c56ea20bb7a78fd5769aae5d0f12b720. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating s8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f1627fbb5b81a5df542173347c1dabb3e3493c579f2d1ec962fbfb8bf88936a7.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 207e717b6d247cfb221fdbcec7a626708ce5b740bdb21f409d1dbbb439941c5e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping s8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f9dcb133309ca12ab7c5fabd0703c014103473203370652bf89f8b5c39de9898.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b95b07cec0fa97e908c16d96d9ded6697b9173ef6debe699b291aa890a5d22ee. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating s8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 35ffdccb3d2495814cf57898071b7b916a6b6ca5627fa62e179523389496a18d.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 619c473b81d7936bbd80cd8b1195a5911b756c6209ce85beba9dc3b63c929df0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 48f8afefe6dcc736f9b1e6750a9aa96bcbe0f1406d48a768f5ec04527b9f8ff2.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e41e8f8421b3066da06e8b9f77e52d19c248e2833be81e57cbe83b1dc4e3b34a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cfc39a65a680d1cbbc4c17da9f23149ff569b977c427d1b56d9e22ae71752d6d.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ed9c523935c6e4d65c559d3b3246acf09a9515e0fade43ca29a79225e3009374. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact wrapping u4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c8888ee61f060e895057d138bf3de9ec113eb39bb2c874055eda73abac1316e9.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 770bae1810b48f7e160fb5a12fc8c49e4075d0c9cf30402bfa707c38346aa36a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-s4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 7,822-byte combined public-order inline-PTX fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) covered all eight m16n8k64 variants with selectors 0 and 1. Pinned LLVM emitted 5,081-byte PTX (SHA-256 d1c8848ea43659574ab6881b36791fa812f2c672b010e37cf735b83a999cd0e8) containing exactly 16 standard m16n8k64 instructions. An auxiliary 6,540-byte typed LLVM fixture (SHA-256 9e11fe0ab804292311e9d13d90a5acb1e8478b044b4110868f7ae5774e762e21) used A2, B2, C4, metadata, selector source order and emitted 22,042-byte PTX (SHA-256 837238a344ec6c789a2c63b970466e299da430cebebcd8dda7b6ab56426c633c) with the same 16 forms. The imported record has ten i32 arguments, four i32 results, ImmArg<arg9>, IntrNoCallback, IntrNoMem, and Range<arg9,0,2>. Pinned LLVM rejected selector 2 (487-byte fixture SHA-256 964fe7dd3cd81b441fd4d38cccac420cfea6d219a27208edf408171f26a97c81) and a dynamic selector (510-byte fixture SHA-256 fc193d79e8e3cfc487cd1aa3b4c312c08efc8d5c34998c3221eba7f423480908). Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted selectors 0 and 1 at PTX 7.1/sm_80 (233-byte fixtures SHA-256 2fdee0d588d2b4f587e59da2ad4e5778a19c69e887017ac331d9cd359b19c794 and 197f1324281ada10b16a5dbd4f65c13808109180b50d29cc5f4524dd849017bb), issued only the ordered-metadata performance advisory, and assembled the full LLVM output to a 4,008-byte cubin (SHA-256 512144584ed413fe89362c698e35b2b128fd144f02b4fa70a26161ae314f203a) retaining exactly 16 IMMA.SP.16864 instructions. It rejected selector 2 (SHA-256 c2d7a5528fa41ef8d04c8e900da46bc41d5cc73e1764c0d10c60aad2e2b6730e), a register selector (SHA-256 d4f972b7c70e68d2b62c86aa181d2eecc73f3bf10e89c570bc478db51218bcb5), PTX 7.0 (SHA-256 d88934934d5bf9fbda0910bd70eadd600e98a4b31bba50e090361dd821bbc90b), and sm_75 (SHA-256 19bf3cb2fa6b19a1e7c1c4cc62c6195b3c8360ebba5785cbc89d46377650c1fa). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7d2ab75eca2756a0c4d2c5544fe7837b92630aa42748c757bf68e0dcf6cc0c8f.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dea3c30a40b3bf4e28518ac19ce412aed32cb5f0394d48bb75ee09953527c0f4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The same 7,822-byte NVVM IR fixture (SHA-256 ea46feacc7fcd1c5ae768178ec7ad5cb97a85c892ca6317fbe4a13c72dafbc33) passed CUDA 13.3 libNVVM and produced 4,979-byte NUL-terminated PTX (SHA-256 abe6ae9238f6137de91c9c2214fc9d7aa78da2bdd6d60a75013f45866c60c38f) containing all eight variants with selectors 0 and 1. Every standard-metadata nibble must encode two distinct 2-bit indices; 0x0, 0x5, 0xa, and 0xf are undefined behavior. This record retained exact saturating u4-by-u4 PTX. Production cargo oxide emit-ltoir generated_intrinsics --arch=sm_80 emitted 374,720-byte NVVM IR (SHA-256 d8180177c4f4e54c641636f7d894fa01314ca6c9251a423ef9027e1985ae120b) and 104,032-byte LTOIR (SHA-256 9f451bb1d66eb39596d0639eb95cf2712e82975f518cf89829eaf18fb5fb15f0). Its 25-byte sm_80 target manifest (SHA-256 06ecfb1b0100fd9886f26871a9fc48350b612a35370b154b07d83a3a57fbbadb) and 49-byte compile-options file (SHA-256 069d630cf5359ec49a571d9074d106aeda207bc279693d2a9ce3c21b168ad03a) pin the production inputs.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — After removing libNVVM's single trailing NUL, CUDA 13.3 ptxas accepted the 4,978-byte PTX (SHA-256 7b24a91e649c24cd57c20d6923a0940730b9465107178b227921b8d5f0a478d2), issued only the ordered-metadata performance advisory, and produced a 4,008-byte cubin (SHA-256 109d906a6082234019e136181d47d7306c65edbedf5c47e3c46ef56635fa6ca6) with exactly 16 IMMA.SP.16864 instructions. The PTX 7.0, sm_75, selector 2, and dynamic-selector negatives are recorded by the paired pinned-LLVM evidence. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — The direct CUDA 13.3 libNVVM-to-nvJitLink route produced a 4,008-byte cubin with SHA-256 3cc463ee7fdde107bbd51039c7a2259f62196923d049c78703e3662cccc378f5 retaining exactly 16 IMMA.SP.16864 instructions. Production linking of that LTOIR through the repository nvJitLink helper produced a 30,952-byte sm_80 cubin (SHA-256 9c9f270e43b9732e814f899d9aeba2dc6f20aef85769bc83eaae8e345cb82ae6). Independent 92,603-byte nvdisasm (SHA-256 dd862d50b376cc5ae8cd083b71d5da500a751b44c4016ab5d6b3c838960cbc7e) and 349,080-byte cuobjdump (SHA-256 44f234c740a4c6f1b3dd917b979a9f495e7e838343130185febc9c604a3e84ca) views each retained exactly 80 IMMA.SP instructions: 20 m16n8k32, 40 m16n8k64, and 20 m16n8k128. The 16 raw and four compatibility calls added by this batch account for 20 instructions. The coverage kernel is unlaunched, so runtime validation remains unexecuted. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping u8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 73597b682d1b12c2fccfad62f40ece013c3c3fcd9024e59c32ef84a8732364e5.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ac63519d007e8dce250d4517871606b252765146f7f59abdd1602790cb331543. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact wrapping u8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c4d93f3a3ddbc1b78370fc6d41ef8514a6a7aabaa7ec338b788f32dd08f910ae.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46b4764da009f8939b74a6235f29b9a6fed892e8b86f68ff56d576e2e5d11151. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating u8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ee679b12117e5c9162717c0d88a751b96da7576d08049d06c19e3fb6c6300c73.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 00174476515d29940f2a3bd1efce00363d2005a85306c962ce53c6a9bc84e43b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_m16n8k64_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx71`: `succeeded` — The 5,051-byte combined inline-PTX probe (SHA-256 9ee9ee8e912fe4dd2b4253197b4fd9a90b70be8a506fd60e9be75e6295866dea) used public C, A, B, metadata, selector order and covered all eight variants. Pinned LLVM emitted 21,538-byte PTX (SHA-256 0ffaffff93d55a366ffac2298876adf9274cd109084fcb9194a25ad08dff074c) with exactly eight base m16n8k64 instructions at PTX 7.1 and sm_80. An auxiliary 4,507-byte typed LLVM comparison (SHA-256 9f0a5a680b4fe7c56ea592123f93013d81441cbcafadb159086bd9e175bcfe4a) used source A, B, C, metadata, selector order and emitted the same eight forms; this record retained exact saturating u8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx71`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions at PTX 7.1 and sm_80, issued only the expected ordered-metadata performance advisory, and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Selector 1, PTX 7.0, and sm_75 were rejected at their exact boundaries. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.1 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a3f0063a7b3cacd62f63384b92d8719e8a4697c12df321ff9a1d5dece0e84d13.
+    - PTX assembly on `sm_80, ptx71`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ff140019dd06530e11bd9507f0e371f07c5d88300c8dbbcd619681437d175825. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.1 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 269,916-byte compute_80 NVVM IR with SHA-256 976d4c2cb32144fc4c6c9923096685fbe1f34306bdcfd2b9b706764508183b78 and 75,912-byte LTOIR with SHA-256 bf01957e2400a1ba5a82246a85b584c1964536bf529a581b64ef4192e2b8a157. All eight raw variants and two compatibility-path calls use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 25,192-byte cubin with SHA-256 c1f9df611c9a113b3830ee6f837232382b95f104c4928668af27a181162afebb. Disassembly contained 40 IMMA.SP instructions: exactly 20 IMMA.SP.16864 instructions (10 new standard calls plus 10 existing ordered calls) and 20 IMMA.SP.16832 instructions. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 5,008-byte combined inline-PTX probe (SHA-256 7145b7febe8ce9f319a106e508a95639d3c06ac7a7cdf1be2c17e0c1f637e516) used public C, A, B, metadata, selector order and covered all eight variants with selector 0. Pinned LLVM emitted 4,025-byte PTX (SHA-256 9eda0f598c87254794f6e3d9e4cae4477b86b5ccb2a0146909c1fed715cce198) with exactly eight ordered m16n8k128 INT4 instructions. An auxiliary 4,800-byte typed LLVM comparison (SHA-256 71c233f005c86cc6616c05c5189171fc62d1525e6e8509927c194177935e8778) used source A, B, C, metadata, selector order and emitted 13,746-byte PTX (SHA-256 5d62e76cc6106809b4a017d82b2b8191dff13107effc9908de13b3a18406eb30) with the same eight forms; this record retained wrapping s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight instructions and produced a 4,008-byte cubin with SHA-256 bdab25ef1571d327ffd4bb0023276418b0f44548683778242827997a726ac660 containing exactly eight IMMA.SP.168128 instructions. LLVM and ptxas rejected selector 1 and dynamic selectors; ptxas also rejected PTX 8.4 and sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7797f8b2f85799bd268bf06f71383c5e441791fb25d6c11afe368dde7b688f2a.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3ddf20d448ac238027732e5de25eceff5461797fae6dadac35f3ff3a32a85a89. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 322,508-byte compute_80 NVVM IR with SHA-256 bb497f219e2fbfc53236897bc82e98ddaf4744ec165e0d00fb5fa729b9131edf and 90,188-byte LTOIR with SHA-256 e1a4cb0426ff501a2343035a9919afcba0621af35511f9d596862bcb728f5c39. All eight raw variants use selector 0. An auxiliary 5,008-byte combined probe with SHA-256 7145b7febe8ce9f319a106e508a95639d3c06ac7a7cdf1be2c17e0c1f637e516 separately passed CUDA 13.3 libNVVM and emitted 3,925-byte PTX with SHA-256 dff59159b2f46e214feb32f6aa60bd64989e75e30cb49da16cc62600654c887f containing all eight forms.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 27,880-byte cubin with SHA-256 f228004e5ee3be5d09bcc6e9c4f6be4a4e81123be335e929f71e3f718d328a81. Disassembly contained 60 IMMA.SP instructions in total, exactly 30 IMMA.SP.16864 instructions, and exactly 10 IMMA.SP.168128 forms: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 468c19abe84ce7bca229d73c81d4ded6564c8d148b95f351a057e04b1b80098e.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8fb8eb364963656a48995d4052deca74cec8dfd07395d74788b843d7264a7128. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating s4-by-s4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f1dcbdef6c39dbf4dd3542b151d9ca227d9ba45ff1e57d7f5d5ee7a93f09a0d4.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cc679e863a7f3653d5ef77a16942b4107959b879360a5b4cf0dd0c3d237cc3f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained wrapping s4-by-u4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5696e1a6c73da853e281aa34b7d6d2fe6dc8be97c211b86c2896a53e6526045a.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75ac82bc1ef628d35241da2b78df5701adbe94efd1da970a2c40b47deb921515. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating s4-by-u4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 943b9548d42ec57747affb74e4de43cd2f8b6dffcc84b3f57c61c014622672e4.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 46e7ba5d6bd8e53adf639c09a6f053a9d08edb608b10649a80c737ced39afaa4. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained wrapping u4-by-u4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained wrapping u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2029ebf4dd778b25d084af4e1383c8952caf0bd2f0c8e5940d2f82be7158e34d.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4b266bce7d640560951da1ff3eda5e0cc1550c8614b2a1ea6183519c8d722903. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained wrapping u4-by-s4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b8cf58c826726dd7c44c2c15f2530af1e33cbe393bd65e8de867ef14ed2276c3.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a19272331578b9876a758d37c2e72dcac32cc7151aa0b2aeb549388f74c8cb51. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating u4-by-s4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with selector 0; this record retained saturating u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 193a37484a4f7f25aa08e90c23783e2bbacb3cac45721e59fc60bd86d6b12c79.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b9455f8cdc76ec4612f947ce914f722c74032563c11075dad9caf80353c14623. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating u4-by-u4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 27,880-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k16_f16_f16`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k16_f32_bf16`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k16_f32_f16`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k16_f32_tf32`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k32_f16_f16`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k32_f32_bf16`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k32_f32_f16`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `mma_sp_ordered_metadata_m16n8k32_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6fd14bda0a191ba64fe0a74811fd81acef22dae9a7364c814feb5dc6b82f25b6.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33a756465fa9b33d89f2b1ab0da1d5ce7a1f0e8293d97888b8db3e9cdc7a1f4a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0097b025872046cef28efa3c9f463bf8993bb9c41da22afc8477efc2a17a4ac3.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 75f59d33a1fc63c903fefc73ee06a31dfa1364d3a5dedcaa5cda933152cc92d2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 64edac3534d432072fb0cb9ee94ae758f539d41f98556515b3c8269967feb7a6.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9111691371ac05725c5828e7342d57b2c0dc17a189490d5fe06a4e499c657703. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b93a06888a9596b56f9ec7dec0e30eb7a87767aa6c4bab12aac88b0c80e4c7e0.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6fb8d4d5c0e41f500f6c523bd132db310bb25628a68ae9cb1331f0a890017d87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cd7e55a90134cfc35729cd87003a44c454b151e8910822841d1d18f0c9ebc1a2.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1566d75ad4047bc212b72c1d4e10c883b8616bb4276963654b1d5a2086c2aa8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12003882bbbb7dab7f4bec48ce3a009d0b9d08e185f69fb277afceef476fb7c9.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e0029151f55c5f3974fcf5c52523d22289796d7bfbd8fa74916f3cee0a82459d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c10be1e6711b2a58cbe3fa9cf60de15556b9496ba056edee6036a3c20b8dd2d6.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5c07f0d6c2eea06bdaa939cf307a9323910485971dd341374537c4067524534c. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The selected LLVM 22 probe retained exactly two unguarded ordered-metadata instructions, one for each selector, at PTX 8.5 and sm_80.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted both selector forms for all eight ordered-metadata INT8 variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c30b514d66dd4ef7d3898ed560a1144be98a172ef526ce4388beaf1f7c7dbd44.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 766d57b5690df60d31209316dc540cfc2f8c31c303b3b3fae6d523385f764f40. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 214,910-byte compute_80 NVVM IR with SHA-256 1c3668d8f8d712f3ea880fded2e10a7e9e4f4265f24f738f104d5bde8e7aee6c and 61,044-byte LTOIR with SHA-256 901eed79871048a57271f434b99bab16cccd7169be6629d6cdb0498bb8f95ece. The ordered raw calls use one compile-time selector per variant and exercise both 0 and 1 across the family.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 22,632-byte cubin with SHA-256 ae24b75fec8d2c8d668765c2513a34c6d707f3acef48d6e106cf256028d72c26; cuobjdump contained 20 IMMA.SP.16832 instructions across the standard and ordered coverage. The ordered half contains all eight raw variants plus two representative compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13684,16 +13684,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13708,16 +13708,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13732,16 +13732,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13756,16 +13756,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13780,16 +13780,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13804,16 +13804,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13828,16 +13828,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13852,16 +13852,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13876,16 +13876,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13900,16 +13900,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13924,16 +13924,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13948,16 +13948,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13972,16 +13972,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -13996,16 +13996,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14020,16 +14020,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14044,16 +14044,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14068,16 +14068,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14092,16 +14092,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14116,16 +14116,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14140,16 +14140,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14164,16 +14164,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14188,16 +14188,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14212,16 +14212,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14236,16 +14236,16 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — The 23,732-byte typed source (SHA-256 ce6f6e73ffd41f5f6c6cf658ff41ce903b7c90afaca38a6279281d2ee8ff60e0) confirmed the imported A4, B4, C2-v2f16, metadata, selector signature. All 25 declarations canonicalized to nocallback nounwind memory(none), with selector immarg range [0,1), and without convergent. Optimization retained only 25 of 50 duplicate typed calls, so this is comparison evidence and not the selected production carrier.
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 18,084-byte all-forms source (SHA-256 cf0e3082d15b64340dab3480ede99adff7d8bada7b8cdfec2fa91e990562ca5e) to 8,991-byte PTX (SHA-256 fc0b67574e5ed2b3cab9a6cdd97d86cd0574c4704f35b0fb73ebaea5acbcb725) containing the exact 25-form matrix. The convergent inline duplicate fixture retained all 50 calls.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2); its 27,724-byte SASS dump (SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270) retains 25 unique QMMA forms. PTX 8.6, sm_120, sm_100a, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_120f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 08692523a26e16fd8a05a2646f71a3ade9e95aa2c3f41028f870d7c9bcab999b) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_120f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 90a0b9bf02df8ca3cb34f867b29b56a65d79a5c9033ea9dc5084ee210b5f6f1d); its 27,701-byte SASS dump (SHA-256 80831cd9a7cde8fd67343302591ea08a30327606719fb232d8e28e812d9c5e46) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121a, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 afec2d0be50c3984867f5402585ba0dbd4f3fe3ac74e6527ab47628ebefd2be4) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121a, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 acd74cced6969aaf6f09afb935d33978ed2753a7f1558240dc4607027b15d502); its 27,724-byte SASS dump (SHA-256 16087533ef8f20e8804dda13de37ea8fce10b505d82ca7c93c5d9d484b0b26e9) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - backend codegen on `sm_121f, ptx88`: `succeeded` — Pinned LLVM emitted 8,991-byte PTX (SHA-256 fae27977151c805ee3bc805a88241f4d4318f0575c558106ed9ba6a0a94f0699) containing the exact 25-form matrix with no unresolved calls.
-    - PTX assembly on `sm_121f, ptx88`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 19ad9907fa56d1a1404de062e1f7dc12b4f350ecbebf57ec8f6066eeda0bfb3e.
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d76b0e1545a10cec1a2b5271a1e7452cf7c9338100ffbf99ec71ca618431f59f.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0d24002860e242490a1617db5a0d1d2915fb068c04b19f2a89a5906ffa15581. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4a0c8f295a64cc1ecde5783756346c9b18e7905c168a615adc561cbc978fad3a.
+    - PTX assembly on `sm_120f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9f20b23371d0816f81e6fd93333205fc9f7f51efd426ad4484fcfca58938354e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1e0fcc258a2f0c2c5a181faf61116dcb63ee8bb85a4db718a69646aeec59aec.
+    - PTX assembly on `sm_121a, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 61fc0290f75f2a55c7422fdaf4a8772869735b72ff43984a1c99d2a97925e1c8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - backend codegen on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 13acdc81de54656db4df63483ce3b0e3a27b679bad246bb22192882bffe243a2.
+    - PTX assembly on `sm_121f, ptx88`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 115294ab5b3a317f4978aa65942dc9feb2768f583896b85c61120f0a2b9cfcc2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 8,886-byte PTX (SHA-256 fe613b565fd6028eaaff9b7df078c93af5b47163f19a9685a43bd6dee28fea6d), produced 11,504-byte LTOIR (SHA-256 aac8c70c26474ccd692e45a453d9f7236c8e040d66dd47047b582524d9deff4c), and emitted no unresolved llvm.nvvm calls or extern declarations.
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 560446ab6b41e45ca208fddc2c1439d9e851b1b67a9de920e6e81026863352b2) retaining all 25 unique QMMA forms; the 27,724-byte SASS dump has SHA-256 3c27689195cd558cd3df06db75ba6e46b1905d39e636536c0a52e01ba3323270. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
@@ -14260,1133 +14260,1133 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 9,728-byte cubin (SHA-256 501b5715323e83c1543a76999f6fb747c6ce1ab0be09aeb637fb95ce69cb28da); its 27,701-byte SASS dump (SHA-256 677ac56dab06986c7ec8938dcc82e23d3637bca8f58282140d0fbce8f483d353) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_121f, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 9,856-byte cubin (SHA-256 7dbcfbe36d9fca5c8916689289754c0041ace50971ff0f9a337c649bc326e2f5); its terminal SASS matches the 27,701-byte ptxas result and retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_120a, ptx87`: `succeeded` — Pinned LLVM lowered the 28,839-byte all-forms source (SHA-256 5518d5dfb092f5ae3ccefe1e7e66eb4df9db7da480722e390587490c6ce8832a) to 12,015-byte PTX (SHA-256 0d93ea01d39bc512b96becf2ad799522d9d5187c17d521294eecd4e69c055e66) containing the exact 25-form matrix at sm_120a and PTX 8.7.
-    - PTX assembly on `sm_120a, ptx87`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. PTX 8.6, sm_120, selector 1, and dynamic selectors were rejected. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be0daf0aa0c20cb20c93aee47dea9157ad8da9e7f39173fe201cda1a731bb60b.
+    - PTX assembly on `sm_120a, ptx87`: `succeeded` — Revalidated 25 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8f1919ac2311c88d71b4dfbacc3ab46cc36448959cdf4456e83bbff0220bfd37. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.7 / sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_120a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained the exact 25-form matrix in 11,910-byte PTX (SHA-256 a096494f6b066927107cf222f784d1a586ed1cf3ab766669b741e429f03f4427) and produced 14,052-byte LTOIR (SHA-256 99941daa24a78884788ed0908a3764f596c527e6ab621d6085c549aea9575377).
     - PTX assembly on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced a 12,144-byte cubin (SHA-256 7769d4fe1de8cdf6644b911bc3654ab220f4dd28cc44a83e28793a5d3f381cf3) retaining all 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_120a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 12,144-byte cubin (SHA-256 62457debbb99023b09b3701e14974a97cfc5761f931ebf9f0d9e2bada3e3a642); its 38,172-byte SASS dump (SHA-256 34df448cb5d7367b36657270961981955bb76cc1dd5dd221f673f174c6e10b0a) retains 25 unique QMMA forms. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 8,107-byte combined inline-PTX probe (SHA-256 ce31bb4c5d7f083ae8e1293a0a3e5196575482427a041d549f9ecd3862e46942) used public C, A, B, metadata, selector order and covered all eight variants with both selector values. Pinned LLVM emitted 8,453-byte PTX (SHA-256 7a8bb0261fd16c1941b47500d70f3cff95a1acb95c51e52480af15a4effc8818) with exactly 16 ordered m16n8k64 INT4 instructions. An auxiliary 6,948-byte typed LLVM comparison (SHA-256 379488c94c4e3d6b1806928d72723845b7bd329857f4dfc3caf8f74d7a5c6649) used source A, B, C, metadata, selector order and emitted 22,330-byte PTX (SHA-256 913b74cc0dba68779655073b83e5a534bcf076fa5e6a1d1d6029ef951b0fc1cd) with the same 16 forms; this record retained wrapping s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all 16 instructions and produced a 4,008-byte cubin with SHA-256 a6df23e0492e6f26ede3da7f7fab6c9fdab2693c5c2d6bda12339948cba640ec containing exactly 16 IMMA.SP.16864 instructions. LLVM and ptxas rejected selector 2 and dynamic selectors; ptxas also rejected PTX 8.4 and sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 34df0473cf3a71f266d58df0459e952a3e93d223f7635e3c0d81fdec210e8e0a.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 884148218b4dbe52217fe5aaceae1dbe2a6322f13e2861aba575e5041cdcb472. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 294,949-byte compute_80 NVVM IR with SHA-256 787025e167bb7bb790fab0af002f838502536834145304ce8df3fc75227059d2 and 82,648-byte LTOIR with SHA-256 eb6516ac258f3c70d20879de097822b13ab73e00288691ae9a13e17bc7fe0d1a. All eight raw variants use selectors 0 and 1. An auxiliary 8,107-byte combined probe with SHA-256 ce31bb4c5d7f083ae8e1293a0a3e5196575482427a041d549f9ecd3862e46942 separately covered every variant with both selectors.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 26,600-byte cubin with SHA-256 1837a2742ce837ccf16b4900b50a72fb018a49bd1850bafe62b3a2a31f550406. Disassembly contained 50 IMMA.SP instructions in total, exactly 30 IMMA.SP.16864 instructions, and exactly 10 ordered m16n8k64 INT4 forms: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating s4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5ef0b5eecdcc862411ed22bc193ddbcea8d94eb7d7f89158a7764a9759137fbc.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 782d840202c153255ac2ac63fdee9b5284b1fb992ec0e82e4fb388e11ba09780. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating s4-by-s4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s4_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e12664cd3c2c8ab1ebee94128e90959dfd9fde36f4023cd59ad4ae6b317a1107.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14dcc3ca9825b8567bc0ab8f60df13f8a003993172ec9d8b6463f31eb12b5b1e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained wrapping s4-by-u4 PTX with selector 1.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating s4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 940e92ff27f876255377efa918c901fed7e3e68e5ea77b987c1e267dc681f894.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d4b51c1e1828a81d1cf6c7fe583ca851ab460871807158be6476b77382dedf9e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating s4-by-u4 PTX with selector 1.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping s8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 978ee747ff8cc5c08d711a0c53853888946906e2669cc24a21928ad8adf71e8e.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 018e9891184767f6e465cb1f24794c6c0145ef5077c2beeca95e42e05aa4189d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating s8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2ae77445ff0a92724257c8a3fb55f6db72266289f6b97e6b47d21516466e5821.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e6527d661a654f0d5d7b508d529f9008ea0aa02aca95bc47712d9669b74a4988. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s8_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping s8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b74c85eca5faf76b32dcda134941e5b09e74cf1aadbfca395a09ad0a4f66286a.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 041f6695d70e0ae239ca5a008247c1c84c651ecdff45cb9a578559f6c0a27731. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating s8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8c147cb0046175cd363c824b78ace3b1acdd06bb5c9d49990db97a715a2ef530.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3859169c6918f08e624d5ace5c0745797f3757d9f96cad14b37655e71e37e405. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1a94dd1d6973c3bba56fbbb5e2972c81d74f9afa874e0e6435de228453839aa4.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b66dbbbbdb29f1a7d3e5cf07a95446ba1875fee261bb0b3e37206aa5c95c9e4d. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained wrapping u4-by-u4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u4_s4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained wrapping u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 206e31e80c020079ad30648d80a62afaf3564e99b9fc60bce10a6d013fe6c3a2.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 22fa426e26b6e51a27796d155fc1fe63e70a16bf4b36d73020d02c786769cd5e. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained wrapping u4-by-s4 PTX with selector 1.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating u4-by-s4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f81cee09a2ed0f8c2a26a84a608b4d8363fb89c6da62fbc53ef36e4669d56432.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb9ce5c9f7687237a762d8ad90791c121859cc83620ca6ef2755a0ead7e93138. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating u4-by-s4 PTX with selector 1.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The combined inline-PTX and typed LLVM probes described by the first record each emitted all eight variants with both selector values; this record retained saturating u4-by-u4 PTX.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted the 4,008-byte combined cubin and the selector and target negative boundaries described by the first record. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06528543fb56e5d6e216226520c60e3fba9c10caee88988d660dc3bf62cfc402.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69d5cab5865f1616c640a859d6acfd3d704c8a4b6a0504edd360b513973809e3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production artifacts described by the first record retained saturating u4-by-u4 PTX with selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — The 26,600-byte production cubin described by the first record retained all eight raw variants and both compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping u8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4aa071df474c2e447be1941f74662bbb1420c01e4fdab1c071aca9d571863b6f.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ca07c7e22bb41deec0862f637e40df38b910149068f4abc8a1e480493439e7f3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u8_s8`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact wrapping u8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 72e7739e891781c3475cf8cacdf67a739e6623660d0f83794818612cd297a82e.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a7361b01c024ab1c79cbe8744f497eaeba5eacbe7ead475ba79e21b317abe01f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating u8-by-s8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 309f99b05e8c94a7d67043662d247b6d2181ace2fa180f341d891848fe49df87.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 1677a6dc99df6612f9de3109d52d45b0293847849900d5ed579f651df96031ea. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — The 4,779-byte combined LLVM IR probe (SHA-256 01600853ff696df81a849a3d804086265d07d61c7a2f24ebaf7b3f056753fccb) covered all eight variants. Pinned LLVM emitted 13,390-byte PTX (SHA-256 b370bcbebbe761bd65675ad2d2602230cc315189a1bb8cb1c4533c68b564a897) with exactly eight ordered m16n8k64 instructions at PTX 8.5 and sm_80; this record retained exact saturating u8-by-u8 PTX with selector 0.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted all eight exact instructions and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Because every probe function was unreferenced, ptxas removed their bodies; this proves assembly acceptance, not retained SASS. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b3d041020ef9ee6fb6013cc491495ec4add4a035da239b9d560b1017e0daeda7.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c2b1813cc3f516d79a19fa139c7fc5d119fc1392db6b1f299fb6ef0ef46df6ca. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production generated_intrinsics route produced 242,464-byte compute_80 NVVM IR with SHA-256 fcc2fe1a6200de7b5ab8a38cab847432669fe4d6ce823c81b73f9b18fa25144c and 68,600-byte LTOIR with SHA-256 aa011070e20cb91d99f0d6d41ffd3883f01031b44d171380f638e3674b2facf3. All eight raw variants use compile-time selector 0.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 23,656-byte cubin with SHA-256 8a09455f87f3ef0c7c16adf13bf6030cbe7f0f18df7fcd91ec2a19a10de5b50d. Disassembly contained 30 IMMA.SP instructions in total and exactly 10 IMMA.SP.16864 instructions: all eight raw variants plus two compatibility-path calls. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mma_sp_ordered_metadata_m16n8k8_f32_tf32`: runtime `executed`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.5 / sm_80+: status `executed` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx85`: `succeeded` — Pinned LLVM 23.1.0-rust lowered the eight per-intrinsic probe fixtures at sm_80/+ptx85 into 59,900 bytes of PTX across eight files; their concatenation in lexicographic id order has SHA-256 2eaae1b352ef07684549050756eba1ae0eb57318cfa329a9f103f086296f81dc, and every fixture retains one unguarded exact ordered-metadata instruction per selector immediate (24 across the family). Typed declarations of all eight intrinsics fail selection at sm_75/+ptx85 and sm_80/+ptx80, and select at sm_80, sm_86, sm_89, and sm_90/+ptx85 plus sm_100/+ptx86.
-    - PTX assembly on `sm_80, ptx85`: `succeeded` — CUDA 13.3 ptxas accepted every selector form of all eight ordered-metadata sparse floating variants at PTX 8.5 and sm_80. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - runtime on `sm_120, ptx85`: `succeeded` — Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.5 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bd546b327c07ec5bef3ec18aaae524027f3ae277b1458336261c3789792dd8f5.
+    - PTX assembly on `sm_80, ptx85`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e469f240cea224beb9f2e7221d32f4ef7db7ce9f7ea859b7b1501aa9b710daf2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - runtime on `sm_120, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: succeeded. Re-measured with the asymmetric oracle on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 8.5 / sm_80+: status `executed` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — The production emit-ltoir route (cargo oxide emit-ltoir sparse_mma_ordered_float --arch sm_80) produced 159,170-byte compute_80 NVVM IR with SHA-256 b338c017795f39f98f756f1d87dc202d1db274479c6619fef9797cb5a95f7613, and CUDA 13.3 libNVVM -gen-lto compiled it to 37,468-byte LTOIR with SHA-256 ac6d8317ba99d9c4332a0ce8777024bc066a7304c6ed4b64c341f4248f6b6100; both artifacts are byte-identical across repeated runs, and the module carries all eight ordered raw calls with one compile-time selector per variant.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — nvJitLink rendered the LTOIR as 83,805-byte PTX 9.3 for sm_80 (SHA-256 4a92814b0579bf19a0ab8aa4c8b2114ec2170c87711831b2a17a31d579bf6e87) retaining all eight ordered forms, and CUDA 13.3 ptxas assembled it into a 41,896-byte cubin with SHA-256 26aa4d2f38132bc6a219ec7b22655a77b412176f464fe93d6c8ae81ea1b9e88b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink linked the real libNVVM LTOIR into a 41,896-byte sm_80 cubin with SHA-256 99aa489c3c80ed1414df0c6fe073cff99dcb66a55fd0057729322b54790ac1bf; its cuobjdump SASS contains exactly eight HMMA.SP instructions, one per variant: 1688.F32.TF32, 16816.F32.TF32, 16816.F32.BF16, 16816.F32, 16816.F16, 16832.F32.BF16, 16832.F32, and 16832.F16. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
     - runtime on `sm_120, ptx93`: `succeeded` — Executed through the libNVVM route (NVVM IR materialized by libNVVM plus nvJitLink at load time) on NVIDIA GeForce RTX 5090 (compute capability 12.0), NVIDIA driver 580.173.02, CUDA toolkit 13.3 V13.3.33. Command: cargo oxide run sparse_mma_ordered_float --emit-nvvm-ir --arch sm_120. Result: exit code 0; SUCCESS: all 8 variants; all 32 lanes and 4 logical accumulators/lane match host GEMM; positionally distinct A/B/C, all six ordered metadata codes (tf32: both), and a nonzero selector per selector domain executed; compute-sanitizer memcheck reports 0 errors.
 - `movmatrix_trans_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_75+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_75, ptx78`: `succeeded` — Pinned Rust LLVM emitted one exact movmatrix instruction; source SHA-256 b8776ed95d8c8e451deea02d8851a238cfcfe570e286a92c94e5971bf06db963, bitcode SHA-256 caf25ab44ed893d449343707f06af85395e86f41852c32ec1b577be184365aee, and PTX SHA-256 f1e1b7d8ad879abd120d6cb32a5999925af0ed77b3a7144a7dfd8db76ea18246.
-    - PTX assembly on `sm_75, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 cbb5dbdfabb866accc3b6dd469ada00f0296aaca7012e3ae14c9b9396da5a887. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_75+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_75, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4d774d39425c57263a42ec917b671796b75c2e755a81255dfc72a40232e32367.
+    - PTX assembly on `sm_75, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 85c6f2304703e7ee71bd8ff7bbb2dec8730ea51244c37934a0cc838b2d00ab69. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-movmatrix` at PTX 7.8 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — CUDA 13.3 emitted a 776-byte PTX 9.3 module containing one exact movmatrix instruction; source SHA-256 b739bbbdb128536727c6d7fe1b9a68475d290b17890620b93723b5ba1f86f7c0 and PTX SHA-256 fc7891162851c2656b68e3a7d5570aad52c06d75ab7ea1a864ab66256783053a.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 produced a 3,240-byte cubin with SHA-256 5d975d8240f586224c39bc2599939ea724d488a0b5c4944e955c13025b1a3bcc. Tool `/usr/local/cuda-13.3/bin/nvcc` reports `CUDA compilation tools, release 13.3, V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `6b55d91cd23f314ccb6801d71d64d20760a7ec31d37d87017ea0681d79bd3285`). Artifact `cubin`.
 - `mul_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact mul.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f4e083050e3dc987284bceea5b0064f86a0c78bca3754d8a92284911bbd7fe78.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6f92a98210d4280598f8059312562b774c73bae8a2baf9af8252031d8f5ac5b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact mul.rn.bf16x2 route at compute_90 as part of the combined three-form probe and produced 1,904-byte LTOIR (SHA-256 87cd4ca7b61dab71aac8885db949e21fbcfbf3bb59ca560c67084d7901ddf60b).
     - device link on `sm_90, ptx93`: `succeeded` — nvJitLink produced a 3,832-byte sm_90 cubin (SHA-256 b0715cda56457dd1b90115ad52c73c8bb920df6f0f4e08df60410cfb4a57635a); cuobjdump showed HFMA2.MMA.BF16_V2 for add, HADD2.BF16_V2 with a negated RHS for sub, and HMUL2.BF16_V2 for mul. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0642183f45157615161325bf3fd8c344f4d8efea483a8bf2c111fd9bd7c9dc4f.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 090bddd2dbb7c5459d46de7c6956de9084a097a5041086b13a2a85322673072d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_75 as part of the combined four-form probe and produced 1,932-byte LTOIR (SHA-256 edb864f3999af5c06558f9dde87cd506ac0c2188cb6a71b19fbf8b33d6b5e5ae). The instruction floor is PTX 4.2/sm_53; compute_75 is this libNVVM profile's floor, and compute_70 is rejected as an unsupported profile option.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,368-byte sm_75 cubin (SHA-256 5ca3d1ca1d9c56802bb78bdad8f5e1138b17dc4dc9f96a7b1e9572eb4010d50c). cuobjdump SASS contains HFMA2, HADD2 add/sub forms, and HMUL2, matching the LLVM/ptxas path. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `mul_ftz_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `mul_rm_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rm_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rm_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rn_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rn_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rn_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rp_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rp_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rp_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `mul_rz_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 44 candidate probes. The 36 selected typed routes matched their imported signatures and property sets. The normalized 220-row artifact index has SHA-256 24dbac0280031cc90cfba25238d09c3194e6616076947b0dc189b5dcccc24126; the canonicalized 44-manifest bundle has SHA-256 f8795846613e0f0c86ff773f3c3bde7047f014dff27c93b61eec301c0b65afba.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The sorted catalog-ID, instruction-head, and operand-count index has SHA-256 743c7c92120bdd6348f8e873929cf343b469b87935d9b857cb61c40dcf672b13. No mixed f16 or bf16 FMA selection appeared.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted every fresh sm_80/PTX 7.0 output. A physical audit matched all 220 payload hashes and sizes recorded by the 44 candidate manifests. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 43a540ca3395340b4c26ec20f43a99ca1e3ae000e3a2093b7b8fc75d9d79eb9d.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3464a16f85a7e5b739d5431bfc395f4c4bab660fd1f2dd69cb07aff3cc3d03f2.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 36 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 33b20a29f610a5c4588eee9fd60186c17523f8362e82e206a4d2e4c5e28ccecb. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-arithmetic` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the 9,966-byte typed-pointer NVVM IR fixture (SHA-256 1887e2d19e9253874b9cc04aa9ddf6fecfe626021185f23c6023da7ec55e1d6f) to 5,956-byte PTX 9.3 (SHA-256 d84a9f54933f0a6d231f81bca8b68939dab13ff12be30ab09f823107dbdce574) and 5,908-byte LTOIR (SHA-256 d11a9ec3ee80833f14fbf0aa68710a1a1e4135a9314f5b7978313ffe6b442ce2). The PTX retained all 44 exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the retained PTX into a 25,000-byte sm_80 cubin with SHA-256 8d204a91958f0abeeaa94e65dda13f8a3238c941ccfca69f81df16ef1dae42f7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and a 25,000-byte cubin with SHA-256 8a820550050fdd6239eea0a9cc5dc0573926fbb49dc58c94ec226e96c8e21f6e. The cubin exports scalar_arithmetic_inline with a 0x4a00-byte text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `nanosleep`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.3 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx63`: `succeeded` — The pinned Rust LLVM retained exact nanosleep.u32 at the route floor.
-    - PTX assembly on `sm_75, ptx63`: `succeeded` — CUDA 13.3 ptxas no longer accepts sm_70; it accepted the otherwise unchanged PTX 6.3 module for forward-compatible sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.3 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx63`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bb1f49e5307d0e617ee6cf95da7ec7a582167ac87e14e8bc4d24cb4185ee7f68.
+    - PTX assembly on `sm_75, ptx63`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 68ce336a8ad5452c31fbff9377bccbcb5540d81a83b904ec1cf2641f607fb34a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-mbarrier-extended` at PTX 6.3 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the module and retained exact nanosleep.u32. The libNVVM route starts at sm_75 because CUDA 13.3 no longer accepts sm_70.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into a cubin for sm_75. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `neg_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact neg.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 1,792-byte PTX 7.0 module for sm_80 and produced a 3,624-byte cubin (SHA-256 0a1a7b10c48e32f35b3cdce0f228a2fdf5d964570c023f3bbd34a09f1ebb7d9d). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2dca2ee0088dbadc0cb0904d95659a2c0d5e7974da0c5e8aff75f0e290504a1.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 779e29da03300e6d8024f6c98694b6bbad23c127181f299430b3a6b09c8541b5. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact neg.bf16x2 route at compute_80 as part of the combined seven-form probe and produced 2,256-byte LTOIR (SHA-256 a9394a9ba0813c36f65bfc2636d7e80b27ab2ffcb6ded14a420d4127a824028d).
     - device link on `sm_80, ptx93`: `succeeded` — nvJitLink produced a 3,624-byte sm_80 cubin (SHA-256 1c8ff2eeb9e58f8b3defc22ed929cd48ec2e2f2beab30747f937a9a7869d74d3); cuobjdump showed HFMA2.BF16_V2, HFMA2.BF16_V2.RELU, HMNMX2.BF16_V2, sign/absolute HFMA2 forms, and F2FP.BF16.PACK_AB. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `neg_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained exact neg.f16x2 inline PTX at PTX 6.0/sm_70. An independent generic v2f16 selection probe emitted neg.f16x2 at sm_53; at sm_52 LLVM emitted scalar sign-bit operations and no packed neg instruction.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the 737-byte PTX module at sm_75 and produced a 3,112-byte cubin (SHA-256 0e381d181a5d707df9d2437d636d47bf1f80cdff578b38160e3553bec0bc870a). SASS realizes packed negation as HADD2 with negated half lanes. The PTX 5.0 negative module was rejected because neg.f16/neg.f16x2 requires PTX 6.0 or later. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e44e1921be6e1f3bc1cc0e885622b5bb8a552876f0f300d6cff83701c2b0a8.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a6abb77861c39cdcaed05cd8512d1b2ca13a32c045cecedb72aa4038fc9c0d6e. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted exact neg.f16x2 at compute_75 and produced 1,648-byte LTOIR (SHA-256 6afc7e94d869c48c78e58d734543f97d19a18099fbe53a572c36b5c22563f10a). The instruction floor is PTX 6.0/sm_53; compute_75 is the CUDA 13.3 profile floor.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,112-byte sm_75 cubin (SHA-256 3e27b5f25b07a5ef425ba7c3e545bacaf1625550ca047db62974c1fb5bf7abf1). cuobjdump shows the same packed-negation HADD2 form as the LLVM/ptxas path. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `nsmid`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75`: `succeeded` — CUDA 13.3 libNVVM verified all six intrinsic declarations before compilation.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,316-byte LTOIR (SHA-256 57b3175e40f2a217ede37a7b8dc3aadce172260ec9ed5459eca6b2d7023711bb) and exact 879-byte PTX (SHA-256 dfac22a9564622fa9b2e9ef64f97a1a50e0e23c853f9734e2aa7c5b8a22ad8eb).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 276fc6a71465f9db688bafa30cd6f85084f291f9fab7a7155e090aa3db1173d3; cuobjdump retained clock and device-constant register reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `nwarpid`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as and llvm-dis produced 2,238-byte canonical IR with SHA-256 a0a8fceff003d4b7a3b3b06c39903cfe73ebaeb3b439126d3445a2a65cded6f0.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted all six reads in 972-byte PTX with SHA-256 cb4020930644bdd8ecf4de1f01416710cbea39f4d098b3a7edd607a8a2a57d65.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 PTX and produced a 3,240-byte cubin with SHA-256 f79ee0584ed55d4083aeec88051cf5e3af4a017fa66d54c51b26dfb4319bed25. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d021bcac1b815c03814998743deebf8b1774364e9b5c597742cc2db1b996b385.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fdf595cddd071a3685c1a510fc1e2de1ca41ec12ac21794bec957c2dcfc34deb.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 6 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 bef9fc2e631cfb651de4e7a61efff21a5bbe614cfb749acf0ff955ef86e4ece6. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75`: `succeeded` — CUDA 13.3 libNVVM verified all six intrinsic declarations before compilation.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM produced 2,316-byte LTOIR (SHA-256 57b3175e40f2a217ede37a7b8dc3aadce172260ec9ed5459eca6b2d7023711bb) and exact 879-byte PTX (SHA-256 dfac22a9564622fa9b2e9ef64f97a1a50e0e23c853f9734e2aa7c5b8a22ad8eb).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 276fc6a71465f9db688bafa30cd6f85084f291f9fab7a7155e090aa3db1173d3; cuobjdump retained clock and device-constant register reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `packed_atomic_add_bf16x2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The selected Rust LLVM backend emitted the exact native packed bf16x2 global atomic-add instruction.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the exact bf16x2 inline-PTX instruction for sm_90. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e2537a05bf31f004570ad178a6290d201b70855af40a531e7bb3c4930907b72f.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5d3e002fed008a798ac7c23848acfa14f680ffa01f2d4c11154a6b127f3b71ce. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx78`: `succeeded` — CUDA 13.3 libNVVM accepted the exact native bf16x2 inline-PTX route at compute_90.
     - device link on `sm_90, ptx78`: `succeeded` — cargo oxide emit-ltoir produced 2,760-byte compute_90 LTOIR; link_ltoir -arch=sm_90 produced a 3,968-byte cubin; cuobjdump -sass showed ATOMG.E.ADD.BF16x2.RN.STRONG.GPU (and the sibling F16x2 instruction). Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `packed_atomic_add_f16x2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.2 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx62`: `succeeded` — The selected Rust LLVM backend emitted the exact native packed f16x2 global atomic-add instruction.
-    - PTX assembly on `sm_75, ptx62`: `succeeded` — CUDA 13.3 ptxas (whose supported floor is sm_75) accepted the sm_70/PTX-6.2 backend output when assembled forward for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.2 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx62`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7769e680b674aa15719338a21a264d9e6a8bd0e7f4a3f90192b4f5507f2472cd.
+    - PTX assembly on `sm_75, ptx62`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3a37572911b7e3d97e3785be11b99068c4b530eac6b0a16d6d74d6a0c6a56702. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx62, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx62`: `succeeded` — CUDA 13.3 libNVVM accepted the exact native f16x2 inline-PTX route at its supported compute_75 profile floor.
     - device link on `sm_75, ptx62`: `succeeded` — compile_ltoir produced 1,920-byte compute_75 LTOIR; link_ltoir -arch=sm_75 produced a 2,984-byte cubin; cuobjdump -sass showed probe with ATOMG.E.ADD.F16x2.RN.STRONG.GPU. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `pmevent`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-debug-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-debug-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-debug-control` at PTX 3.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 emitted a 562-byte PTX 9.3 module containing all three instructions; source SHA-256 2e8b8ffd062afdcf1c1806d526388b9bacc7469918370d26af9186b2c9a4b1e1 and PTX SHA-256 ba2d42ec37ae06d4a460291a966872cbc35e999ffd0aa72c7e1aa6cf0f54e6f8.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 produced a 5,416-byte cubin with SHA-256 d9c0af472531aa6da4a84efc0b4cbaccc56f9e090c5c3160792842172c225ee1; cuobjdump retained PMTRIG and BPT.TRAP instructions. Tool `/usr/local/cuda-13.3/bin/nvcc` reports `CUDA compilation tools, release 13.3, V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `6b55d91cd23f314ccb6801d71d64d20760a7ec31d37d87017ea0681d79bd3285`). Artifact `cubin`.
 - `prefetch_tma_descriptor`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — llvm-as and llvm-dis canonicalized every declaration. SHA-256 over sorted per-operation ID/hash lists is 699f97e8c0cf8e87e56866abea35827690b76e706f28d99b4031469d80f3c596 for source IR, 87defb3cdfce6aec545000fa473129442b0fe93db210ac3acba69a8897796162 for bitcode, and 5bd87c1b69729bd627d2ff4b0ffccfc417c12b643405732efd52a35a8a2eea1c for canonical IR.
-    - backend codegen on `sm_90, ptx80`: `succeeded` — The pinned LLVM emitted every exact instruction at its catalog floor; the sorted per-operation ID/PTX-hash list has SHA-256 428d749128dd3f204dc06a4c6d66a9cffb5227ff2a7314e08031bc7b7cbabefe.
-    - PTX assembly on `sm_90, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 011d7563ffe5319b17237455e156d3edfd77c028b5cb6965fa4a785f68074521. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.0 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 445b3338f4553b6b1a95af6425ce02988353b46f94664c1eeeff1f8ba2a1e714.
+    - backend codegen on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 44ca85cf203984b118014ccd762241fce1c0761edf13ba904c62beec51fbf466.
+    - PTX assembly on `sm_90, ptx80`: `succeeded` — Revalidated 11 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e9a9f694e7471f96d44191df80bcfdcffdc1dfd2afb54ae5af72de964ae7fc0f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.0 / sm_90+: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is bc9032cec56c22c29dfe509c5525129b9c9bc93d8ae1b476faa31fae766e24fc for source IR, a046fa221b8a2aecb62d154de8e73d937c87398b5b0da91929fe8dce837af4c9 for LTOIR, and 400058b1c886ea9a2b68870c57ff1beed7290291cb6f74aed83372e4eafb6630 for linked PTX.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 aa75eebb5ce7527516063375ce098561ea5ce2f9dc481cfca20eb9f7518d0f1a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `prmt`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `prmt_b4e`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `prmt_ecl`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `prmt_ecr`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `prmt_f4e`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `prmt_rc16`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `prmt_rc8`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — llvm-as accepted the canonical attributes; IR SHA-256 0ee6196e1f67ae7ee42660cee813a5cbe5b0d247dbdbca079ef4559f26c324ea.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — LLVM emitted all seven modes with dynamic controls and all seven immediate controls; PTX SHA-256 765f7cb84c57545ee6f09cd5fedc27b7e4bca47153f4dd468bad18f1855d6fb8.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the virtual sm_20 floor PTX; cubin SHA-256 0d714af03e6771b42bfbb0cb2a8f81e645aa81e12a829c414d5ffd90ea27a9e7. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a084bf24c4d2d075cbcaf304683a38225ca216fa5aa4ce5b6253f4a612ca0be2.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69dca80d2402aa1c98601ceacff7551390bee53445425729a9274b7b10f1c5f2.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 7 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 30a5c98c7cbcb52630c62c15106c39139b280fe0716e20772486aef675458b72. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-prmt` at PTX 2.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted every mode; LTOIR SHA-256 82beb73a1fde7677a85801ad07a59d8949aa05c36111d7c632af171b3d0703d5.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced PTX and cubin; PTX SHA-256 dcf052adef81afc5af5769e4495f7527929e7ae604631583aa61de75486cc609, cubin SHA-256 2c1eb32f40834c3ff328d02148e68da3ede02610bb6bc39979c1b45756d3efdc. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_approx_ftz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rm_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rm_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rm_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rn_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rn_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rn_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rp_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rp_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rp_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rcp_rz_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rsqrt_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rsqrt_approx_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `rsqrt_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `setmaxnreg_dec`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 8.0 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90a`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 a559d3e87b535201a9ae87b367b6bbc3cd6c6fc452574311161fa126b2cabfb7, bitcode SHA-256 6a87d7b67460deeb9fca1720371d2d9f1295620546fc167c64deee03d362543b, and canonical IR SHA-256 b5754b28a81283e193437f849a7e07af5488d484fcb5c3c3d2e9a2d15c193d2d.
-    - backend codegen on `sm_90a, ptx80`: `succeeded` — LLVM emitted the exact instruction in PTX SHA-256 9758b60ad9ed87b7e0a90f9fd57c3edaa90c7c793621c5f911f0fcaef0a55555.
-    - PTX assembly on `sm_90a, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 8.0 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90a`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 be9be58f6dcddd987a8632ca75a2ed6a592f6a65ceff960adc24bcb52bb11db3.
+    - backend codegen on `sm_90a, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddfec7d31a47990bd677ada57d94afc3bf77325e79e886219aac161fffc0f896.
+    - PTX assembly on `sm_90a, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 37ad4f1284701ac5101858b0792ed16e833420d4e21cbc1c02896d03019a63a2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 8.0 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 b9d4789bf7c7e6265427203eaf0d2fd2b2c0e2b7e5af33b27ec0ff53debf157f to LTOIR SHA-256 1b7d8499637583aeaf1954612605a4278ae1e29ef4cf1823928cf238b049a733; nvJitLink emitted matching PTX SHA-256 8828781cfc87509545585b74eb5338a745d1120454e798a819f9a61b7308f8f2.
     - device link on `sm_90a, ptx93`: `succeeded` — nvJitLink accepted the instruction and produced cubin SHA-256 09fe97fe245219bd114da9e46252415a83d870d45ddde64e79fe380ea2bfb960. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `setmaxnreg_inc`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-execution-control` at PTX 8.0 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90a`: `succeeded` — llvm-as and llvm-dis canonicalized source IR SHA-256 e451448a7385c879e0ca127c25cd038b92bffeacf94d6c3487dbeebbf4f2555e, bitcode SHA-256 00e92b75a1ef024a48fb31297bd9a46d831be64e916d3a522f16bc54d19046a3, and canonical IR SHA-256 7d3888169df2d273686c57ac7b169aaee07b0917f6db7b148a6403130aba3a68.
-    - backend codegen on `sm_90a, ptx80`: `succeeded` — LLVM emitted the exact instruction in PTX SHA-256 1212a8cf1cbba526030c0f67dbb854c5dd2d24e7cf7977ef2137a14638be0d8a.
-    - PTX assembly on `sm_90a, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX and produced cubin SHA-256 97ff03a8de0c3a6044a5bd7b909640d72ff6b0136c9843bc395ca7e16cfd8f95. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-execution-control` at PTX 8.0 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90a`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c6648953f506cce13f94612bf6d0da7cd3e075fbc63a6afc01f2ae709fb9a415.
+    - backend codegen on `sm_90a, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4032051215c86d96bb88330e799aec22fb09ad3d232fe483e2346692108ff967.
+    - PTX assembly on `sm_90a, ptx80`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 06608466abf4fe979cff6b7e5e3a16a92d0daf5df071239d28e943cd7be29916. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-execution-control` at PTX 8.0 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — libNVVM compiled source IR SHA-256 5df660b3f34a8c45d1ec9a8bca38024084bfc025bbfc414179db9a753a11d557 to LTOIR SHA-256 75b77b6e2edb6cbb43c8fa19f80be297f73ff3a29b956049e5419d9b87b35c77; nvJitLink emitted matching PTX SHA-256 ebf19d2fcde31e23f608880f61b2c4d5b916a4cc58fd87235e74918c0b6badde.
     - device link on `sm_90a, ptx93`: `succeeded` — nvJitLink accepted the instruction and produced cubin SHA-256 632a69e158fb616805f1aba4fe87fa7e15372e60f12ef26f6ccd3add16d20dea. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `shuffle_down_f32_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.down.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fe998b216b523ddba326d894f7fd6c1c811b7b8f6793fd67fa9f32bb5426fe93.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e822e31752229f9d5f8379f58427407f4da67974fdf5cb800973a1d420d99843.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 78a64562571e8dfa2d02c779947163cd4c087bf0fb45cfc9b26b7a9448da1d34. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_down_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.down.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 12f77b50bd9a2ac2e167e43862d4925890afbc944ce1be9abf7a42bef834a693.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e868aa2aec22540ec973c73eda5ad85c82e432f562a4956b89fc5ec681ff12c4.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 56902b3d564c19bb5e6ffca02b1f0f73c0c6ea1a440d7ba209fe9242e86d6b50. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_down_u64_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 8dc004e784b4ad1d4151d0fbbb8c7ab4815437672e35cb37c8408d685938d961.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ac475b6fa0fb52beb3890ebaba00ef224041684fd47bfec82f609768fb0ad71f. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the reviewed low-half then high-half recipe and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 4,512-byte LTOIR (SHA-256 ea189a7d375862e72996f49f95b4f660a1c56c2edaf3c7d39e58b953e1e9d1b2); nvJitLink produced an 8,936-byte cubin (SHA-256 860fcf7329acaa4e8a77e00860b1c89f3e6d671e9d008b8901464bf4eb5263a6); cuobjdump showed all four shuffle modes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_f32_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.idx.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d782063e4b1ae122d58089206151ace8d5b7d4f660a5d5159fb1b770f56b46a0.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e4775aad2f0dbe6b4f3fc2db7d15985eed28a8adcc2031100d3db5a4c25600a4.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 578376a2dccd0ed45a721f981ae2bf7094d65d90e46479f5569dc8a00354275f. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.idx.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4c8015a293ea78a1394796bab968f8d4b065b433149894cf07283122ab75837d.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 dcc151e0a9e33ce7a059a418221504ddf54bb43d6f1b8927cffd7cdd76c93939.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0dab85a67df2645eacc4979c404b7be148188a35e648c7e180b82aba4269ae71. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_u64_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 00eed2a3ce51403c06b4557ca323d691094cf59685823dd73b0b329a2fc40d9a.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d1301d100cbe1985f544b748d01b5a92e32173c169be106074f0614f23f7a08d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the reviewed low-half then high-half recipe and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 4,512-byte LTOIR (SHA-256 ea189a7d375862e72996f49f95b4f660a1c56c2edaf3c7d39e58b953e1e9d1b2); nvJitLink produced an 8,936-byte cubin (SHA-256 860fcf7329acaa4e8a77e00860b1c89f3e6d671e9d008b8901464bf4eb5263a6); cuobjdump showed all four shuffle modes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_up_f32_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.up.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5a97a689d300dd7c199c714c01189afb154122960104cfc1a1f1e1180c4e7cf3.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a3ed93208637ba99cf7cdbdd4ddbcc7dd13f263fcecc0842c46753c8c841ca34.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 01942dd79db0ae31a8cb526a0bab0e1bbac0a00e9494f7fd2c1ac989af6a68b2. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_up_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.up.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d6fef706af3d2993713934e1f5156c47628a6c943899ad369edf140f7b9f5c57.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e9ac34403d0a172c008ea25d9e62a0786926818eec2efb2091811ddca6c0caf.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 faebc8c202740b059817f861bd3cded8188e653457a4b9791ce87506a1c63314. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_up_u64_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a83c401208ac85d4c6fe20f942147f8bd8410442b4ce60b86079b138ceac2208.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a85d3dccb30f9491380b1f54913e7261c58667a5c22f195298c6a4f0738f3992. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the reviewed low-half then high-half recipe and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 4,512-byte LTOIR (SHA-256 ea189a7d375862e72996f49f95b4f660a1c56c2edaf3c7d39e58b953e1e9d1b2); nvJitLink produced an 8,936-byte cubin (SHA-256 860fcf7329acaa4e8a77e00860b1c89f3e6d671e9d008b8901464bf4eb5263a6); cuobjdump showed all four shuffle modes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_xor_f32_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.bfly.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fa450ce0f0bf68458c30deaa762636db4ffcfbd33d8f8c5d627a057b59740efb.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f67d1796c2477484afa60ecc1be1bc640a045537ba52e1512f1cb17bc46898e4.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4ab01d57f019c3f6994b977e6528bbadebb9db1980ff1e4e57480354c0cb1aa0. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_xor_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent, nocallback, and inaccessible-memory attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted all four reachable shfl.sync.bfly.b32 operand forms with the fixed clamp.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted every generated register and immediate form. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cc205d40d08b4db407d73012b4e02bacca951cbf2407578220e1a9993241f7c4.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6d1ab9e1df5d932129b785eb98a099581f503fadd7a7e22b4e7621f06dd4ecce.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f4465879f688e4239d1fc44f8e1c2008923c890cf59e0d1e2cad6653348622d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — The pinned LLVM 22 probe independently established the modern typed declaration's convergent, nocallback, and inaccessible-memory attributes before cuda-oxide legalized it for libNVVM.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA-oxide rewrote the modern typed call to libNVVM's legacy aggregate shuffle intrinsic, preserving the mode, value bits, lane or delta, and fixed clamp. CUDA 13.3 libNVVM accepted that IR and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 5,944-byte LTOIR (SHA-256 bb1bd56ee46b4bce228e99e2e2cfac099d39a64c497c0a1cf87615dbf294c7be); nvJitLink produced a 5,160-byte cubin (SHA-256 ce2b2ca9ac1d39f4833c0549a743a238db490d820a6bc8f5f51a9fa89be77a38); cuobjdump showed two SHFL.IDX, SHFL.BFLY, SHFL.DOWN, and SHFL.UP instructions. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `shuffle_xor_u64_sync`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted the reviewed low-half then high-half recipe with register-only lane and member-mask operands.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the exact two-instruction recipe when assembled forward for sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a423fb03456e022c274ab0aee6f4480239603e12b7657f78a3cce23264b020e6.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 41a4b0b1e404dde5ac1a6f457e13bde66b3a26ce037c64f3c691fa4666f07d5d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the reviewed low-half then high-half recipe and produced LTOIR.
     - device link on `sm_75, ptx93`: `succeeded` — libNVVM produced 4,512-byte LTOIR (SHA-256 ea189a7d375862e72996f49f95b4f660a1c56c2edaf3c7d39e58b953e1e9d1b2); nvJitLink produced an 8,936-byte cubin (SHA-256 860fcf7329acaa4e8a77e00860b1c89f3e6d671e9d008b8901464bf4eb5263a6); cuobjdump showed all four shuffle modes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sin_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sin_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — These 9 intrinsics (sin.approx.{f,ftz.f}, cos.approx.{f,ftz.f}, lg2.approx.{f,ftz.f}, rsqrt.approx.{f,ftz.f,d}) have valid LLVM NVVM declarations with IntrNoMem properties but no imported DAG selection patterns. LLVM lowers them through NVVMIntrinsic class pattern matching.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted the correct PTX instruction for all 9 routes within the shared 37-function proof bundle. The PTX output has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the shared proof and produced a cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 848e863721c151f4c3bdc1ae80df9bedba7cc7edd2c3d680399119821e67b082.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 094fb972e1487a21ff367ca7549357317a976326625e1ebbe4ae71f490cbc03b.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 9 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 832391a539356a083929e9c33a555ed3d0b71a4ddde88e08e8541fca97689540. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `smid`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted both volatile location samples in 633-byte PTX with SHA-256 8b6bdea0bf90c68baecd7189ae84ba5c24a0ca32a3a347a1dc72ca8a0a62277f.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas produced a 2,984-byte cubin with SHA-256 2a1f9febb72d01384c5c1c4583e4511020b2c086144d8b25f6aad354a4c34618. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ec007869a74a8d4d9bf7be114f276c2c4e26f3e4c34eeaab721362bbf9a1918.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14f47ee4cc2165c26a56b268fb647e325de01ae94e284257de734b97dea0626a. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 1.3 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained globaltimer, smid, gridid, warpid, dynamic_smem_size, and total_smem_size in 2,028-byte LTOIR (SHA-256 e00f26c9c773f5ea24198cda4a679a0de54ceafb3a13be5190b4dcf0b475f864) and emitted 1,135-byte PTX (SHA-256 4cb7846fcc9a10b3d26ccbab61c1936f26ae6255911d637cf6875fbf4e6d3bea).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 60456548b55c17dbef4e6a269c3eacb20298af44406ed81eec5657362501cdda; cuobjdump retained globaltimer, virtual SM, grid, warp, and shared-memory size reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_approx_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rm_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rm_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rm_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rn_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rn_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rn_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rp_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rp_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rp_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rz_f64`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sqrt_rz_ftz_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Pinned llvm-as and llvm-dis canonicalized all 37 candidate probes. The 28 typed routes with DAG selections (rcp.{rn,rz,rm,rp}.{f,ftz.f,d}, rcp.approx.ftz.{f,d}, sqrt.{rn,rz,rm,rp}.{f,ftz.f,d}, sqrt.approx.{f,ftz.f}) matched their imported signatures and IntrNoMem property sets.
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed direct instruction exactly once. The 37-function PTX proof bundle has SHA-256 70a1594b2887bdf4f67fbbc45570545d6045fb1b5822824caa532175fd7557d3.
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas accepted the 16,065-byte sm_80/PTX 7.0 output and produced a 1,464-byte cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 fea78dbe8828029a051c62bc546d364a72b197a6b98a1b47854322807aefc125.
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 872fd820a42be92ef628a5cb1f79f46895142f0acb65cc5cd978d0bcee090536.
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 28 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 40510f96011426cd1c7d1ca35cbcd48ae2decebcd625310fd37db6fc047b8f7a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the typed-pointer NVVM IR fixture to PTX 9.3 and LTOIR. All 37 scalar math variants (sin, cos, lg2, rcp, rsqrt, sqrt with approx and rounded precision modes) are retained as exact, unique instructions with no extern calls or unresolved llvm.nvvm symbols.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the 37-function PTX proof into an sm_80 cubin with SHA-256 6bcec1b1eab13da703fdade053704aa6ab00591a1b38dd7daf48ac0dec541df2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink emitted linked PTX byte-identical to the direct libNVVM PTX and produced an sm_80 cubin. The cubin exports scalar_math_inline with the expected text section. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `stmatrix_m8n8_x2_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-stmatrix` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the 1,544-byte source module with SHA-256 fe4f9eae0f68076b87a64db475a3025703b5fba3c3594234601c0b491bef5adb, emitted 1,582-byte PTX with SHA-256 f72734d526a1de3477f7a681c6918b9310634223dfcc447ab644d0825cecf95e, and produced 2,828-byte LTOIR with SHA-256 1eb129c866ca475e500c0218d85b818bb63890b9dd09026da86d8082e887af51. The PTX retained each exact instruction once.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into a 1,728-byte cubin with SHA-256 690c0e13b8a66020ebde091bf13e2b421cb191e6e2a1a04531e120f9d773912a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `stmatrix_m8n8_x2_trans_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-stmatrix` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the 1,544-byte source module with SHA-256 fe4f9eae0f68076b87a64db475a3025703b5fba3c3594234601c0b491bef5adb, emitted 1,582-byte PTX with SHA-256 f72734d526a1de3477f7a681c6918b9310634223dfcc447ab644d0825cecf95e, and produced 2,828-byte LTOIR with SHA-256 1eb129c866ca475e500c0218d85b818bb63890b9dd09026da86d8082e887af51. The PTX retained each exact instruction once.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into a 1,728-byte cubin with SHA-256 690c0e13b8a66020ebde091bf13e2b421cb191e6e2a1a04531e120f9d773912a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `stmatrix_m8n8_x4_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-stmatrix` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the 1,544-byte source module with SHA-256 fe4f9eae0f68076b87a64db475a3025703b5fba3c3594234601c0b491bef5adb, emitted 1,582-byte PTX with SHA-256 f72734d526a1de3477f7a681c6918b9310634223dfcc447ab644d0825cecf95e, and produced 2,828-byte LTOIR with SHA-256 1eb129c866ca475e500c0218d85b818bb63890b9dd09026da86d8082e887af51. The PTX retained each exact instruction once.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into a 1,728-byte cubin with SHA-256 690c0e13b8a66020ebde091bf13e2b421cb191e6e2a1a04531e120f9d773912a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `stmatrix_m8n8_x4_trans_b16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all four declarations. Canonical IR SHA-256 values were d1f71c7c75af39dccc3b8bc5da08960ac7e21862d06274c4002bd27aca2b5b83, 1b3144318581f7a5f995bc08332dd8430e4ce15e131b7f5d3f0d11367c34ce22, efd8f41836b357d64176dbf6814ab3a6ca5067f6c7d0be29c7acd337f09d7318, and b2a4b9e0c2b8ea91d7a464e0f3ed8074e00034135760edd995f354ed860745e2.
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The pinned Rust LLVM emitted each exact stmatrix instruction at the route floor. PTX SHA-256 values were d785aeb501608a3245f5c278c9c9abf76e8304a6859e3510f51b671254101a64, 1c42bafa2a28719b52031f9736a310bf7cd85f14246c0416366fe287115866e2, 29941042e09a032655bcd2f4cb2d240e94dcd301cefdc2e1c7d947cfb6348bc6, and bfae167f02c68d9776a8755a3abfd9abc2239619b6763bb26372ebee50304ce1.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,720-byte cubins with SHA-256 f573c8cd30697e3635bd5bdf741092b49232ccd88cb86113d55a74b6f2b23843. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b30e7ee609317814c806ad2b54f4841b23cab9fe5cf72db69c0594488424c016.
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 178dadce496e866eb4707712183af0bc8b9eb4564e0a00be7c5d4f0f5d94f764.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 4 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 efc8374add158b09083ae65e3e1bb77c540742528efd699c5669c81ec5789bdf. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-stmatrix` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the 1,544-byte source module with SHA-256 fe4f9eae0f68076b87a64db475a3025703b5fba3c3594234601c0b491bef5adb, emitted 1,582-byte PTX with SHA-256 f72734d526a1de3477f7a681c6918b9310634223dfcc447ab644d0825cecf95e, and produced 2,828-byte LTOIR with SHA-256 1eb129c866ca475e500c0218d85b818bb63890b9dd09026da86d8082e887af51. The PTX retained each exact instruction once.
     - device link on `sm_90, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced LTOIR into a 1,728-byte cubin with SHA-256 690c0e13b8a66020ebde091bf13e2b421cb191e6e2a1a04531e120f9d773912a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sub_bf16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90, ptx78`: `succeeded` — The selected Rust LLVM 22.1.2 backend emitted the exact sub.rn.bf16x2 instruction at its native floor as part of the combined packed-bf16x2 probe.
-    - PTX assembly on `sm_90, ptx78`: `succeeded` — CUDA 13.3 ptxas accepted the 1,071-byte PTX 7.8 module for sm_90 and produced a 3,832-byte cubin (SHA-256 a498ca833a4c45100c3e80c165999e4d9389d263e47a909a7f97a25c114822c0). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.8 / sm_90+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 379f54089c227f714a7a906db098eef11d0ac042db017cbd022236e2f366c440.
+    - PTX assembly on `sm_90, ptx78`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7e912f52acc5af50229ee7d87a37888ffbfd4dddc236e582d317512c281864d0. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 7.8 / sm_90+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_90, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact sub.rn.bf16x2 route at compute_90 as part of the combined three-form probe and produced 1,904-byte LTOIR (SHA-256 87cd4ca7b61dab71aac8885db949e21fbcfbf3bb59ca560c67084d7901ddf60b).
     - device link on `sm_90, ptx93`: `succeeded` — nvJitLink produced a 3,832-byte sm_90 cubin (SHA-256 b0715cda56457dd1b90115ad52c73c8bb920df6f0f4e08df60410cfb4a57635a); cuobjdump showed HFMA2.MMA.BF16_V2 for add, HADD2.BF16_V2 with a negated RHS for sub, and HMUL2.BF16_V2 for mul. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sub_f16x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_70, ptx60`: `succeeded` — The selected Rust LLVM 22.1.2 backend retained the exact pure inline-PTX instruction in the combined four-form PTX 6.0/sm_70 probe. An independent generic v2f16 selection probe emitted exact add.rn/sub.rn/mul.rn f16x2 instructions at sm_53; at sm_52 LLVM scalarized them and emitted no packed instruction.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted the 1,252-byte PTX 6.0 module at sm_75 and produced a 3,368-byte cubin (SHA-256 3f23db63edf3340a638d464a8423346292f17846bed5e8c921d01c03371a8633). SASS contains HFMA2, HADD2 add/sub forms, and HMUL2. The PTX 4.1 negative module was rejected for every instruction as requiring PTX 4.2 or later. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_70+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_70, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 260eb7ac7c16c07d7eb3ac88e498b18d9e9f37b1c763d3dca78075888ea7cce8.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdc2a8724f570ba9cf569ea9240cf60a937116228c42a5d0d9bdcef1af591d35. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_70 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact route at compute_75 as part of the combined four-form probe and produced 1,932-byte LTOIR (SHA-256 edb864f3999af5c06558f9dde87cd506ac0c2188cb6a71b19fbf8b33d6b5e5ae). The instruction floor is PTX 4.2/sm_53; compute_75 is this libNVVM profile's floor, and compute_70 is rejected as an unsupported profile option.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 3,368-byte sm_75 cubin (SHA-256 5ca3d1ca1d9c56802bb78bdad8f5e1138b17dc4dc9f96a7b1e9572eb4010d50c). cuobjdump SASS contains HFMA2, HADD2 add/sub forms, and HMUL2, matching the LLVM/ptxas path. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sub_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `sub_ftz_f32x2`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100, ptx86`: `succeeded` — Pinned LLVM 22.1.2 retained all eight exact add/sub/mul/fma .f32x2 instructions, including every .ftz form, with =l/l constraints. SHA-256 over the ordered intrinsic-ID/source-hash manifest is a04546b877e8f07b55c5dc11a6a5071651b53bbd15077a43083e59f3e267853a; the corresponding PTX-hash manifest is cb56147f343b2dc89724f6e0cce42826b32589d0e620d5df71725d214619cd6d.
-    - PTX assembly on `sm_100, ptx86`: `succeeded` — CUDA 12.9 ptxas accepted every module and produced valid CUDA ELF cubins. SHA-256 over the ordered intrinsic-ID/cubin-hash manifest is bf1c99f3a81819e7c21fbd79d53da0e7e4045efe25792110d3334a8c37e056bf. Tool `/usr/local/cuda-12.9/bin/ptxas` reports `CUDA 12.9 V12.9.86 build cuda_12.9.r12.9/compiler.36037853_0` (SHA-256 `983b0e9283855979f42cebfd80d43f9b6e786eb84f03f7570bf941c4d3a3c461`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 28abb9f72234f874e80bb09c68101ba626804a5a66e9b4499fc890a8cdf3aae1.
+    - PTX assembly on `sm_100, ptx86`: `succeeded` — Revalidated 8 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7791ae79668395fdb8147d82d5759db156ce6de0bdffe66c1e4f74f5cb50487. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-12.9-libnvvm-12.9.86-f32x2` at PTX 8.6 / sm_100+: status `validated` (`CUDA 12.9 libNVVM package V12.9.86; NVVM 20.0.0`, SHA-256 `7915b9e3eaed7f686774ee82f99cfacebaefcd00296f44bdb979e57d363a7ca8`)
     - backend codegen on `compute_100, ptx88`: `succeeded` — CUDA 12.9 libNVVM accepted the 1,794-byte module (SHA-256 3bb81df41493fd8254071078d5423dcdf5bbcdea542b46a77cf14d4d62081620), produced 3,060-byte LTOIR (SHA-256 623cfd620ace08aa14f84559fe9297b89f032ebac8eee747c38f3f499469e02a), and nvJitLink emitted 1,558-byte PTX 8.8 (SHA-256 8920b773f314dd4a60cdc620d168dd2acd0c6ee453a8f231ec1303f62665355f) retaining every exact instruction.
     - device link on `sm_100, ptx88`: `succeeded` — CUDA 12.9 nvJitLink produced a valid 6,008-byte CUDA ELF cubin (SHA-256 dc9e6034e77cb3fa6d4046a4b6bf5aa5e2b0d804bf7884638b093cd804e2caff). nvdisasm contains the expected FADD2, FMUL2, FFMA2 and .FTZ forms; subtraction appears as FADD2 with a negated source. Tool `/usr/local/cuda-12.9/lib64/libnvJitLink.so.12.9.86` reports `CUDA 12.9 nvJitLink V12.9.86` (SHA-256 `02d3acb5fe598dd20f0fca3cc03734ad164037a22747a01900561a42d0b8448f`). Artifact `cubin`.
 - `sync_mask`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — LLVM recognized the declaration and preserved its convergent and nocallback attributes.
-    - backend codegen on `sm_30, ptx60`: `succeeded` — The selected Rust LLVM backend emitted exact register and immediate bar.warp.sync forms.
-    - PTX assembly on `sm_75, ptx60`: `succeeded` — CUDA 13.3 ptxas accepted both forms and produced a 1,464-byte cubin (SHA-256 df48ead94aa2ad3402456419362350879e99993d4b2ce03a05f695e9d58fbf37). Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 6.0 / sm_30+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d5b275af3bb7cfa86185e4ce735308f68edfb3dd4a71261c8ed80131f8f67b84.
+    - backend codegen on `sm_30, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9e7bf824b7fcf96ef357d4279091d73c4cd95081c9a10994d3a2abc4df06f58d.
+    - PTX assembly on `sm_75, ptx60`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 395cf308dc13aa17943c5d073be46dc163c23cd9add8444b68a0dcff491240de. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_30 +ptx60, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 6.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the typed declaration with its convergent and nocallback attributes.
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM emitted exact register and immediate bar.warp.sync forms.
     - device link on `sm_80, ptx93`: `succeeded` — libNVVM produced 5,212-byte LTOIR (SHA-256 617a67c02ae53343922fa2ab38a8b6d022914f5c1a64c901e3cb79f1ea0511c5); nvJitLink produced an 8,360-byte cubin (SHA-256 11bc13b23e68cb7c86ffc1e1be09037e018509c57b34674c835d9a34d719158a); cuobjdump showed WARPSYNC. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `sync_threads`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — LLVM recognized the declaration and added convergent and nocallback attributes.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The selected Rust LLVM backend emitted exact bar.sync 0 PTX at its oldest supported target profile.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the emitted PTX when assembled for its supported sm_75 floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 25b383a09f3293d7f5ceef1bc66d8400702d7a5c68d73f24532c371030e69c11.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 26b139fc6199abb20c9a7d92ddb0a30839fb2bdd3eafe285ec63bb49da4bb3c5.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e287209ed91d6c476b48a3d08399a0911d64f601f3ae07e328db3f975fc2d2e9. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 1.0 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the exact inline-PTX route at compute_75 and produced 1,656-byte LTOIR using its PTX 9.3 profile.
     - device link on `sm_75, ptx93`: `succeeded` — nvJitLink produced a 2,856-byte cubin; cuobjdump showed BAR.SYNC 0x0. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tanh_approx_f32`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_80, ptx70`: `succeeded` — Pinned llc emitted each reviewed instruction exactly once per probe. Per-op PTX SHA-256: ex2_approx_f32 e5085463d64720c2769ca57b8c1e4eece25b26452b3308cf873ba94d7daa17b2, ex2_approx_ftz_f32 8ef33c541bb966b084a3bea0ee452b3210ae48509d8d1228d4427d895b938ef0, tanh_approx_f32 e25146dfbb77a3ad5b6ad4553b861c0a4bb2a39df8c4a7d87b49f4314fa72c30. The hashes are byte-identical to the original llc-22 attestation (LLVM version 22.1.2-rust-1.96.0-nightly, SHA-256 ff1dcdc18a5441003d43c5df267b43e2b6dd3efccca38c20bb2801885626331e) because these inline-PTX probe bodies pass through llc verbatim. ex2 binds the overloaded selection-less import records; tanh has no import record at all (PTX-native). The typed llvm.nvvm.tanh.approx.f32 route was re-checked alongside this llc-23 re-attestation: the pinned llc 23 and Ubuntu LLVM 22.1.7 select it via NVVMIntrinsic-class matching, while Ubuntu LLVM 21.1.8 still lowers it to an extern funcall, so the inline route remains required while LLVM 21 remains a supported llc fallback. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - PTX assembly on `sm_80, ptx70`: `succeeded` — CUDA 13.3 ptxas assembled each per-op probe for sm_80. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 7.0 / sm_80+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_80, ptx70`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7f7959d76a6bc77d5d1dbd1e909cfd96c0c0c99cc5a5053c4f664f6cd0eddcdb. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - PTX assembly on `sm_80, ptx70`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 d0e13be12bb5695c5107849e200384164007696566dcf5a51f890bee17ba356a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-scalar-math` at PTX 7.0 / sm_80+: status `validated` (`CUDA compilation tools, release 13.3, V13.3.33; libNVVM IR 2.0 debug 3.2`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified and compiled a dedicated typed-pointer NVVM IR fixture holding the 3 appended variants (ex2.approx.f32, ex2.approx.ftz.f32, tanh.approx.f32) for compute_80. The emitted PTX 9.3 output retains each instruction exactly once with no extern calls; PTX SHA-256 9d2a3b28989cf40d76816090ce6051c2347f46d8bbbd6fd7fb5b479977e5d7cc.
     - PTX assembly on `sm_80, ptx93`: `succeeded` — CUDA 13.3 ptxas assembled the appended-fixture PTX into an sm_80 cubin with SHA-256 134a765e19dd2267cf41934ad4ffa3ec0a6f54c43406d4da3102adfb9334e955. tanh.approx.f32's hardware floor is sm_75 (PTX ISA 7.0); the family admits it at the attested sm_80 evidence floor. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.33; build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM compiled the appended fixture with -gen-lto to a 1,848-byte LTOIR (SHA-256 6c6fcf6eb5ddef3cf3ea605c0a866c8cc4db63bae1df7abca521ff507283b56d) and nvJitLink 13.3 linked it into a 1,464-byte sm_80 cubin (SHA-256 3e4a06eab6319f8d75a890f4b2645d35d92125c79eaa2b63dba2e5383552d632) with all three appended instructions retained. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `nvJitLinkVersion 13.3; CUDA 13.3 V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_alloc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_alloc_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_commit`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_commit_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_commit_multicast`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-control` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: compute_100a 0b0bffdde9ac419ddef9aa2dfab0a88d7478e27860e804fcbbfef332c1920754; compute_103a 9019c74a58c6077102fc2223da9f67b52850f4df7ad73cf7784e05009f86b65e; compute_110a e1b1bd804b6201a86a382021b60dbaf141e3aac10dbc7c26f12ad98f80322910. The separate direct-intrinsic fixture also succeeded on those targets, with PTX SHA-256 4c535d8f76a09ef5d2b38f07f6a1f1a4dda5afc4e0540f89b987f7e5a0cb3c4f, a3500c104fce6291d8e1b3badc091085078d499f72b508eb5598cc12d1f30954, and fd38746edac8a01e45d978db36999ca83a1132ba18b4327bfbb54c2f1413f44c, in target order. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a/f with NVVM_ERROR_INVALID_OPTION. Direct shift intrinsics rejected compute_100f, compute_103f, and compute_110f as unsupported operations; the commit intrinsic separately passed those family targets. Direct forms of all three records rejected base targets and compute_120/121. The shared libNVVM route is closed to compute_100a, compute_103a, and compute_110a.
@@ -15394,1137 +15394,1137 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_100a, ptx93`: `succeeded` — LTOIR SHA-256: sm_100a c79c3b541a41faa35ff05353e8aa7738b7db3201c59cd0403fb4f888d622555e; sm_103a 8563b5b7b21a1a4161aa9216bbd707fe22224815ada8a79928d4279cc00fe84c; sm_110a fcccad4be6c5d7e623a49876d7ba8859ab08dff720682d7a3d713432ab2aea4a. nvJitLink cubin SHA-256: 73e1691afa49d02c15f8cb938e9ec1f12c8464cf4f74749f9ffa48cf4974500a; 420c4fd951b26f081a34d73aaa0160acfb4f999fbf5e809bb450eb3b958bda35; ceeaf53f9251c600daef97387490e636b3403e85a51af33133f51cadada21658, in target order. Every linked cubin retains one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_commit_multicast_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_commit_shared_cluster`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_commit_shared_cluster_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x128b`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x128b_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x128b_b4x16_p64_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x128b_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x128b_b6x16_p32_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x128b_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x256b_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x256b_b4x16_p64_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x256b_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_128x256b_b6x16_p32_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_32x128b_warpx4`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_32x128b_warpx4_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_32x128b_warpx4_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_32x128b_warpx4_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_4x256b`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_4x256b_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_4x256b_b4x16_p64_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_4x256b_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_4x256b_b6x16_p32_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_4x256b_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_01_23`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_01_23_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_02_13`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_64x128b_warpx2_02_13_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 5,109-byte typed-pointer source (SHA-256 914baad4ce112754fb0b34f2b6824766eb977e77196fe13ea4fb11c581340e21), produced bitcode SHA-256 23002f057d0d5720ddce06d5def55ef0958af5682300468888b69f01ee8c4d21, and retained all 34 exact instructions. PTX SHA-256 was fab1756b278a3fd00a85797c4c0a376376b043226798a0b4c8a2acbd047d8f6a for sm_100a, b00d0bbc9e4824c51232257204abc8150c53f114c2f1abb6b5d3462071aff6d9 for sm_101a, e64627305ede8083e348ae81af005ff5b9b232cfa19989252121c52fff76c9f3 for sm_103a, and 748958f648295b3300257b90ff15dec54e13f5df372d009a671a0f55df43b908 for sm_110a.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the complete split fixture on sm_100a, sm_103a, and sm_110a, and nvdisasm found exactly 34 UTCCP instructions in each cubin. Cubin SHA-256 was b58967c2e67fa470c1a12a47d01f3d02fc65261958aa69ebfe8958b2cf4b8918, ad6d3268878dab454f4cf67a8d063ec83572a22f4506ca93f357a6c36fe458a4, and 7d2269bc1422d7e57760b7c7762a1fc368980da99a57598932210aec7dfd746b. CUDA 13.3 has no sm_101a assembler spelling. Boundary probes rejected sm_100a at PTX 8.5 because tcgen05 requires PTX 8.6. CUDA 13.3 ptxas also rejected the complete fixture on sm_90a, sm_120a, and sm_121a because the tcgen05.cp instruction, member shapes, or CTA-group forms are unsupported on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7cfd9b8aa83bcac0d97bc736db550472f47781701ac3a98b16e15036bb05d449.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 34 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e8bc112fb0e0c1d0a1526739d0bb1e6c8d399cc7bbedb8dbcf4d020538e6b34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-cp` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the typed-pointer split fixture and retained all 34 instructions in PTX and LTOIR. PTX SHA-256 was 3dfc4f0336bb8045cdb3a203e11ee04b2d9e8f074fb61b8c324c76adfa28c82d for compute_100a, 87c38d36b6fead694fd2248bc74ada93d18bc75b2442f4aa2f0e899ab23266ad for compute_103a, and 3f959712aa1717771aa9526cd1700d3587119c2e3ed49b5e0bcda789bb2b1b3f for compute_110a. LTOIR SHA-256 was 0da88e4b46ab921223df6a9502f9fbc33c1a6b558f7cac026c1ca3a4ca646585, 3be6c1a8b4a9c8b66d5bdacd77bb22d650c8b92bd5215de7bd171042b4e9c53e, and c1f59e1b9f7b4e161798093af0a03bbc4cd1afd040901dbbfbab479d513b4479. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas accepted all three libNVVM PTX outputs. The resulting cubins were byte-identical to the direct LLVM terminal cubins for each target. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted each LTOIR image and retained exactly 34 UTCCP instructions. Cubin SHA-256 was dbf48c876a09e6ccc7e367f4ff43882c9efb5019fd2623150f19cd018dbf9013 for sm_100a, 83072d1f282cb05e4d807cb20a1153c920e0cf8a588083bbc00fd8f457ba4e55 for sm_103a, and b0d84748520ad92264dac257b24da070b11077c38d9637b6d59377d5e40cdee5 for sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_smem_to_tmem`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_cp_smem_to_tmem_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_dealloc`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_dealloc_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_fence_after_thread_sync`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_fence_before_thread_sync`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x16_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x1_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x2_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x32_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x4_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x64_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x8_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x128b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_pure`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x16_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x1_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x2_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x32_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x4_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x8_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x8_pure`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x256b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x128_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x128_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x16_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x1_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x2_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x32_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x4_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x64_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x8_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x32bx2_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x128_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x128_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x16_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x1_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x2_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x32_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x4_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x64_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x8_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_16x64b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x128_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x128_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x16_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x1_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x2_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x32_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x4_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x64_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x8_pack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_ld_32x32b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894). Every target retained 58 entry kernels, 58 exact loads, and 29 pack::16b forms. PTX SHA-256 was 69c0fb5696fb664a099e5120f0d18ad18f00dff1ebd50100ca9645387d98b1d0 for sm_100a at PTX 8.6, b4b66448a03f9b7e3c6c18463c68cf237c68465ed69837ead3474b2bf61c05de for sm_101a at PTX 8.6, bcaa35e58cc69b6d854dbc344463470a1b7a7c0b0aead0d68db1df89ce155b5e for sm_103a at PTX 8.8, and 481483b98b8657f6731442523b07feaa9eef37a1f1b39dc1f21a0be3db3c5928 for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 ee832a6286fde07b52a7800a10911cb756a12793de68daf6fb4e6f94f7578f04 for sm_100a, 7fed294ec249f9e8278a316ba46cd9133bab10a03f1be5d57164307e710849c8 for sm_103a, and c2ca947c990b23f1287def0e60eee881ce5e4264b2c439179ce43f8ff63fc160 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. CUDA 13.3 has no sm_101a assembler spelling; its final-source PTX still contains all 58 entry kernels and exact loads. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 000e155b300b75d3b3ae0f28eacdb6f8bc4844bdf6eaca6f1b92611fcafada58.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6ad93cef3016dbdee49358bd8e3dd5a2612f0bcb42580b26c904343780002499. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.ld, all four load shapes, and pack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ld` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, i1], their exact scalar or vector result, ImmArg<arg1>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg1 to false; each pack16 leaf fixes arg1 to true. CUDA 13.3 libNVVM accepted the 58-kernel fixture (source SHA-256 7d3e9d0c1cd4b46a66699fe5f1a651b55a9151bdb13eb7ad5ed5c605c1e61894) and produced LTOIR SHA-256 259e62b9b8b7167ffaf3bb0653690c48d4ff6a68fa8197ba0d308fa701539f6f.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 b1e5522bd2048d84a379e83cba6af06c8971dc7c47d2aeb97c8b92356ce824bf. nvdisasm found exactly 58 LDTM instructions, including 29 PACK16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_load_wait`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16548,25 +16548,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16590,25 +16590,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16632,25 +16632,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16674,25 +16674,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16716,39 +16716,39 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_f16_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_shared`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16772,25 +16772,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_sp_shared`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16814,25 +16814,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_sp_tensor`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16856,25 +16856,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_sp_tensor_ashift`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16898,25 +16898,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_tensor`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16940,25 +16940,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_tensor_ashift`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -16982,32 +16982,32 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_bf16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_e2m1`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17031,25 +17031,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_e2m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17073,25 +17073,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_e3m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17115,25 +17115,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_e4m3`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17157,25 +17157,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_e5m2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17199,32 +17199,32 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_f16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_shared`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17248,25 +17248,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_shared_zero_col_mask`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17290,25 +17290,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_sp_shared`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17332,25 +17332,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_sp_shared_zero_col_mask`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17374,25 +17374,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_sp_tensor`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17416,25 +17416,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_sp_tensor_zero_col_mask`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17458,25 +17458,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_tensor`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17500,25 +17500,25 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_tensor_zero_col_mask`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM emitted all 24 generated probes at the exact PTX floor: 30,652 PTX bytes in 24 files, sorted SHA-256 manifest c2ad5ba66e4139cc89c14712b870da372cac61d7e85045e931688e9808311065. The four direct f16/tf32/f8f6f4/i8 oracle outputs total 3,500 bytes, manifest 552a6101ef26cb380916a3e2b695f6da24a001f7f6b2a0a1358b8fa430bacab7. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. All 14 forms, five new aliases, and five preserved aliases were retained. KIND and A/B collector selector mappings matched LLVM. Runtime was not executed. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest c0131f21c3da3d37e0b58975d43fd9ec08eb0dbfd12c60a7bdcbd8e12bc07caa. The three supported direct f16/tf32/f8f6f4 outputs total 2,646 bytes, manifest 24af25ad04abb2ecd0241a0a495954ddc6579832411e5b459336949c465fdd10. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_101a, ptx86`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 4ada11e892ab8dfc3ab5a37f53a0d55da1c6c4eac95b284746aa51076b112b5e. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 4c4024ec32c46e397744c0defdfc6f608440b720b04b7aae11107198b280b2db. Runtime was not executed.
-    - backend codegen on `sm_101f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e9e41c78b623b647c135416968582c89e1e84e4d02546d952bf4b3df5d5f2f11. The three supported direct outputs total 2,646 bytes, manifest f271cdae12c7abf1d92dc8ef0189a6b17d732e05eea21e8a4f9a979df7956a73. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103a, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest e56b5d5937469c1cab73321246bf692245f1714d1bc72bdc048edb4decd15764. The three supported direct outputs total 2,646 bytes, manifest f226ffd247e8fd2d5fdbde1e5e0d94dd2f90d65877bbcd14420d56a921cad535. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_103f, ptx88`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest eb93ff0229050788f1daf636ff75c3c0f9e5248e0975063da8553f17b637eb60. The three supported direct outputs total 2,646 bytes, manifest b8913ad71ef46105375af33b7e667c8918adcf0504b0e68bf00846557a16c867. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_110a, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 1572d296738398575115494712243257edd0fe957b64222c72270bd11e6bf013. The four direct f16/tf32/f8f6f4/i8 outputs total 3,500 bytes, manifest 96c8f6c57ef5f6c225fe9326801d1d8217db13e672be8011ebe97fd0c1baaa3e. Runtime was not executed.
-    - backend codegen on `sm_110f, ptx90`: `succeeded` — All 24 generated probes emitted 30,652 PTX bytes, manifest 62319016965629b25e314375eabcca88ee897188eeddbd2a8f68271e25439d49. The three supported direct outputs total 2,646 bytes, manifest 8df9973acec65f0120150aa735b7ab172720058b9e38d1883ad7045052bf646a. Direct i8 selection failed as required. Runtime was not executed.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
-    - backend codegen on `sm_101a, ptx90`: `failed` — Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
-    - backend codegen on `sm_120a, ptx90`: `failed` — Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — ptxas produced 24 generated-probe cubins totaling 53,952 bytes, manifest cda261c8da15355daa6f52fa8a1f9afe93617b45443c6038f9f946717f277bea, plus four direct-selector cubins totaling 21,792 bytes, manifest 56adcdb2321c5852dcfdedf7503f0c8643f20bd31e9321763694124666abd1f6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_100f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest e4629d467551e62e195db21cd88dcf275796ca64681973121fa85def1e28771e. The three direct cubins total 16,376 bytes, manifest 197fbe2b9c37896ff72286893c3baa845c40160babfdafc6288dbc00ef342a28. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103a, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest b5d874681e92ad1d6b13754a3a62bfcaa006fdac2c0f6a3cf6931c1b1f867b8e. The three direct cubins total 16,376 bytes, manifest 272ca391de3edab856a8f0ac050f46de23bf43959b4b17cc73405c8baf121c4b. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_103f, ptx88`: `succeeded` — The 24 generated cubins total 53,760 bytes, manifest 7d8207ea3d6668ba639296ffaca84bc2de642b32c0ea78a4ff2587cb1c5b3ce8. The three direct cubins total 16,376 bytes, manifest f0377b848d177809731224640ef2134c9540dbc0e00c7e08298fc0fd5925cc5f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110a, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest be63b0d33e9548f0d36347caa2f50f7972c486bf9b07d73cc7ea7a7fef7e692f. The four direct cubins total 22,912 bytes, manifest ee63aecae0e05955bfb2449f09afb65a7b68cd10fa50e8a72bc55d1ac2235ab6. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_110f, ptx90`: `succeeded` — The 24 generated cubins total 60,096 bytes, manifest 6bdcbcbdc3b1e15d50a96e50207268194dc8c9f208da6de5051cba33ccd58825. The three direct cubins total 17,184 bytes, manifest 6a5f86031c0fd82b159b4558d2ca8787aaea80bcd6e2cb58c1d69162a1601f78. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_101a, ptx86`: `failed` — CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_101a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_101a at PTX 8.6 or sm_101f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `lowered` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c667e9fca3549231de67ead0b1fd0bca52619f4f1495be421f6228d6b19a82ad. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 16f495ff0a6480322f673a6a02ff256f736eecd89082d22acc8de7a8f818ba69.
+    - backend codegen on `sm_101a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 58392ab343525376265c230b2754c1cb8a5f743da4f87e00c8c7d4a316fbdaea.
+    - backend codegen on `sm_101f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 91b6a9212aefa3d9bae3fdb9ea1a39ac5b7b4b7e38a0a3dbe5e59b31c565a33c.
+    - backend codegen on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7256ef08da331a00012655e4fb9c0fd99edfe6fb9a9444e6cf9c48dba1fda323.
+    - backend codegen on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c1f09ef64d6589d58a5fcd00e9506575ae862fde933908883dc6d7e5ea7daf51.
+    - backend codegen on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ddf2d1f7b71994f94bded64e98c7ffc8f11b6be0e41929397710c5cb180c85c6.
+    - backend codegen on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0290ee182fa49c1284809f6accb1c627e711175244d4fac4d325a21d8f4e55f1.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a/sm_101a at PTX 8.5, sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7, and sm_110a/sm_110f at PTX 8.8. PTX 8.9 is not an LLVM NVPTX feature and was not used as a floor probe.
+    - backend codegen on `sm_101a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected f16 on sm_101a and sm_101f at PTX 9.0 and also rejected i8 on sm_101a at PTX 9.0. The paired target matrix must fail closed when a module requires PTX 9.0.
+    - backend codegen on `sm_120a, ptx90`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM could not select f16 on sm_120a or sm_120f and could not select i8 on sm_120a. tcgen05 MMA is not admitted on sm_120 targets.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2cf4fefa9b98cea1818e43509f8b95ef7dc94590286aebb20c94c1393a112fd9. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_100f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 83135ea6cafa894c5a69675f8264a36c2cca1edc0589f8eb56c34abdf1813917. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103a, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5226c51f90c015caa2b541b6c7e09579ced0821729199c1d1a2e0fd83f2bafc3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_103f, ptx88`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab094dd8c59cbbf994b85636ee0c23854383b53bb40f33964cc1525f921ccd87. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110a, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ede544853c18650f798bf1621cfe54b8d9275f7662bfa6fb7d4587440c52aa34. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_110f, ptx90`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 4f71dec6f3ac32438e5cb6a3eaa8c6071fcf7f975868dcea2da31408f9566395. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_101a, ptx86`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejects both sm_101a and sm_101f because neither gpu-name spelling exists. LLVM emitted all 24 probes for both paired targets, but no CUDA terminal assembler is available, so this profile remains lowered rather than validated. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-mma` at PTX 8.6 / sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f16; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=f8f6f4; sm_100a at PTX 8.6 or sm_110a at PTX 9.0 for kind=i8; sm_100a at PTX 8.6 or sm_100f at PTX 8.8 or sm_103a at PTX 8.8 or sm_103f at PTX 8.8 or sm_110a at PTX 9.0 or sm_110f at PTX 9.0 for kind=tf32: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — libNVVM emitted all 24 decorated generated probes: 28,730 PTX bytes, sorted SHA-256 manifest 016235453738d6088a2997bd5794c956523cb6a855ea034580f3e36d877555dc. Every manifest in this fixture hashes lexically sorted sha256sum entries keyed by artifact basename. Decoration added only the required datalayout, NVVM version, and kernel metadata. All 14 forms, five new aliases, and five preserved aliases were retained. A separate direct i8 oracle passed this target. Runtime was not executed. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_100f, ptx93`: `succeeded` — All 24 probes emitted 28,730 PTX bytes, manifest b887844c3f1b24ece0cc5e70ac9f9481cc641d7e1bc65a95b405667f0aed44f4. The generated probes passed the terminal assembler. A separate direct i8 oracle was rejected on this target. Runtime was not executed.
@@ -17542,32 +17542,32 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — Representative retained f16 and i8 probes were rejected on sm_120a and sm_120f, confirming that this tcgen05 MMA family is not admitted on those targets. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 ptxas V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_120a, ptx93`: `failed` — Retained f16 and i8 LTO probes reached nvJitLink and were rejected during PTX compilation on sm_120a and sm_120f. @llvm.used ensured these were real terminal failures rather than empty-cubin false successes. Tool `/usr/local/cuda-13.3/targets/x86_64-linux/lib/libnvJitLink.so.13.3.33` reports `CUDA 13.3 libnvJitLink V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_mma_ws_tf32`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_relinquish_alloc_permit`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_relinquish_alloc_permit_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_shift_down`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-control` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: compute_100a 0b0bffdde9ac419ddef9aa2dfab0a88d7478e27860e804fcbbfef332c1920754; compute_103a 9019c74a58c6077102fc2223da9f67b52850f4df7ad73cf7784e05009f86b65e; compute_110a e1b1bd804b6201a86a382021b60dbaf141e3aac10dbc7c26f12ad98f80322910. The separate direct-intrinsic fixture also succeeded on those targets, with PTX SHA-256 4c535d8f76a09ef5d2b38f07f6a1f1a4dda5afc4e0540f89b987f7e5a0cb3c4f, a3500c104fce6291d8e1b3badc091085078d499f72b508eb5598cc12d1f30954, and fd38746edac8a01e45d978db36999ca83a1132ba18b4327bfbb54c2f1413f44c, in target order. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a/f with NVVM_ERROR_INVALID_OPTION. Direct shift intrinsics rejected compute_100f, compute_103f, and compute_110f as unsupported operations; the commit intrinsic separately passed those family targets. Direct forms of all three records rejected base targets and compute_120/121. The shared libNVVM route is closed to compute_100a, compute_103a, and compute_110a.
@@ -17575,11 +17575,11 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_100a, ptx93`: `succeeded` — LTOIR SHA-256: sm_100a c79c3b541a41faa35ff05353e8aa7738b7db3201c59cd0403fb4f888d622555e; sm_103a 8563b5b7b21a1a4161aa9216bbd707fe22224815ada8a79928d4279cc00fe84c; sm_110a fcccad4be6c5d7e623a49876d7ba8859ab08dff720682d7a3d713432ab2aea4a. nvJitLink cubin SHA-256: 73e1691afa49d02c15f8cb938e9ec1f12c8464cf4f74749f9ffa48cf4974500a; 420c4fd951b26f081a34d73aaa0160acfb4f999fbf5e809bb450eb3b958bda35; ceeaf53f9251c600daef97387490e636b3403e85a51af33133f51cadada21658, in target order. Every linked cubin retains one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_shift_down_cg2`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: sm_100a/+ptx86 fe1818fb3e1217277df5f8922dc413380c8d13ee6a94f35401182aff0e875554; sm_101a/+ptx86 f78d319d104c813972755fb159f0a3f947c9a28e145af19dae04e79bd4711b35; sm_103a/+ptx88 44e7d4fd8b50b0973a5db7652603703b7bef6981e324fcfdcbaa94a929cb46d5; sm_110a/+ptx90 f4fd13829114b6240d81c6e8eef90994cef310201a4d0c989b9af4f107e35017. A separate direct-intrinsic fixture (SHA-256 5d6f385d95b18564caed76220ac69296e1e625ae6d98c0c48e6311d8df034c21) emitted the same instructions on those targets. Direct PTX SHA-256: c4ae06b9fd3894dc070c1e86c5addb338b9fb6a7c80ab3b15bff961cd5264325; 99b083ffd590e834dfa45f3f5e86ca00b4d98863308bef0912a24d337303c22c; 86671718d828e2082b66c67a4e89ef49abef80186db1b2618b579b7b1b4475ad; 1a782892745df5e6647c6c1827f8b515905d76ec49dfc6f84a563b1faa6f5599, in target order above. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted sm_100a, sm_103a, and sm_110a. Cubin SHA-256: 88743111f57596a0c77bafe054764c63b5dafe8f4a72a49e35aabe41865feeba; 8e50db9432c58adb4ed83badb4984720f7592aeba322ff3b97740e0caf31c486; cdead147e86914c22f5fde7a325f142c2e88f0dc56a8261056937eadeb9bec24, in target order. CUDA 13.3 has no sm_101a assembler spelling. Each cubin has one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 b7368219d0f51e0e65a18632cbf4f38337cb173fe797770469aa6a0c6ddbd8a4. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5, sm_103a at PTX 8.7, and sm_110a at PTX 8.8. Both shift records also rejected every tested base and family target. The commit record separately passed sm_100f/sm_101f/sm_103f at PTX 8.8 and sm_110f at PTX 9.0. All three records rejected sm_120 and sm_121 spellings. This closes the shared route to sm_100a, sm_101a, sm_103a, and sm_110a at their proved floors.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 0ddbfb5241b0d4b9d3e3c84b5b1f7d23f557e4072d83ef074484a26aa31dfb7f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-control` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained three entries and the three exact instructions from the 1,180-byte source (SHA-256 c9d6009791b64f0871bf7d007d85d381ba94d1125070ebcdac4c7f0fcb652e13). PTX SHA-256: compute_100a 0b0bffdde9ac419ddef9aa2dfab0a88d7478e27860e804fcbbfef332c1920754; compute_103a 9019c74a58c6077102fc2223da9f67b52850f4df7ad73cf7784e05009f86b65e; compute_110a e1b1bd804b6201a86a382021b60dbaf141e3aac10dbc7c26f12ad98f80322910. The separate direct-intrinsic fixture also succeeded on those targets, with PTX SHA-256 4c535d8f76a09ef5d2b38f07f6a1f1a4dda5afc4e0540f89b987f7e5a0cb3c4f, a3500c104fce6291d8e1b3badc091085078d499f72b508eb5598cc12d1f30954, and fd38746edac8a01e45d978db36999ca83a1132ba18b4327bfbb54c2f1413f44c, in target order. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a/f with NVVM_ERROR_INVALID_OPTION. Direct shift intrinsics rejected compute_100f, compute_103f, and compute_110f as unsupported operations; the commit intrinsic separately passed those family targets. Direct forms of all three records rejected base targets and compute_120/121. The shared libNVVM route is closed to compute_100a, compute_103a, and compute_110a.
@@ -17587,976 +17587,976 @@ This table is generated from the resolved catalog. Evidence stages below disting
     - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected tcgen05.shift on sm_100f, sm_103f, and sm_110f. It rejected tcgen05.commit and tcgen05.shift on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
     - device link on `sm_100a, ptx93`: `succeeded` — LTOIR SHA-256: sm_100a c79c3b541a41faa35ff05353e8aa7738b7db3201c59cd0403fb4f888d622555e; sm_103a 8563b5b7b21a1a4161aa9216bbd707fe22224815ada8a79928d4279cc00fe84c; sm_110a fcccad4be6c5d7e623a49876d7ba8859ab08dff720682d7a3d713432ab2aea4a. nvJitLink cubin SHA-256: 73e1691afa49d02c15f8cb938e9ec1f12c8464cf4f74749f9ffa48cf4974500a; 420c4fd951b26f081a34d73aaa0160acfb4f999fbf5e809bb450eb3b958bda35; ceeaf53f9251c600daef97387490e636b3403e85a51af33133f51cadada21658, in target order. Every linked cubin retains one UTCBAR.MULTICAST, one UTCSHIFT.DOWN, and one UTCSHIFT.2CTA.DOWN. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x16_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x1_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x2_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x32_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x4_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x64_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x128b_x8_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x16_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x1_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x2_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x32_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x4_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x256b_x8_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x128_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x128_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x16_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x1_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x2_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x32_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x4_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x64_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x32bx2_x8_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned llc accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Every successful target retained 32 entries, 16 exact loads, 16 exact stores, 8 pack::16b loads, and 8 unpack::16b stores. PTX SHA-256: sm_100a/+ptx86 72ed5dfd5d8a2765dccfda9f74cc920c7247292ab0905b4a7cf8d41d49a7e674; sm_101a/+ptx86 19327e2058f9ad42ffc5e5f8d9de74de5c2644e7b02b682d0848d91503b22868; sm_103a/+ptx88 69887c9ce9d2432248ef22837a88a4a3be6cc0b413a34bd5fe4a8938c96a34a2; sm_110a/+ptx90 7df33a256a701f57d97fe1c4c9092d40dd7ae52daac7539ce5d7391e426b6534. Separately proven family-target PTX SHA-256: sm_100f/+ptx88 f1fa2c2ba513ddab0ff041eb610dde14058c5d81847e2d041514403c4df23ecd; sm_101f/+ptx88 d034643204b14dd6345d4ca777cb22bf0bbeaa9e7a3b74b0313ed04a219031c6; sm_103f/+ptx88 71f659cc80f4906a66a803d0d92f608327af514210a9b1e2071c7d09b2baf740; sm_110f/+ptx90 fe44017ddee11dd0cef2562890d9ea69f2113667d21bd5925314d8d33c46de81. The imported records have no selections, so the admitted route is inline PTX. The public i32 constant is sign-extended into the imported i64 immediate carrier. Selected negative i64 direct-intrinsic probes crashed in LLVM's Assembly Printer, so that route is not admitted. PTX and CCCL state no numeric range; the admitted public API follows CCCL's i32 carrier, and the production fixture uses offset 16. Tool `llc` reports `LLVM version 23.1.0-rust-1.100.0-nightly` (SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`).
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas cubin SHA-256: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. CUDA 13.3 has no sm_101a/sm_101f gpu-name spelling. Each assembled target has 32 kernel sections. The 16 load leaves each expand into two terminal halves, yielding 32 LDTM; the 16 store leaves similarly yield 32 STTM. The 8 packed load leaves yield 16 PACK16BIT halves, and the 8 unpacked store leaves yield 16 EXPAND16BIT halves. nvdisasm found 32 t0_t15 halves at the base and 32 t16_t31 halves at base+0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx93`: `failed` — CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 69730742a6b92d3b7dfb1ae54a6c32ab3d8ee3cb5844bb22ed83a1fd20a9c1f6. Tool `/home/nihalp/.rustup/toolchains/nightly-2026-10-11-x86_64-unknown-linux-gnu/lib/rustlib/x86_64-unknown-linux-gnu/bin/llc` reports `LLVM version 23.1.3-rust-1.101.0-nightly` (SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`).
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a and sm_101a at PTX 8.5 (minimum 8.6), sm_100f/sm_101f/sm_103a/sm_103f at PTX 8.7 (minimum 8.8), and sm_110a/sm_110f at PTX 8.8 (minimum 9.0).
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 32 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 e3fa9eaf195b7592f5088445b5bf502293a03b644cb39a7b0937cb79529b10b2. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx93`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected .16x32bx2 and tcgen05.ld/st on plain sm_100, sm_103, and sm_110 and on sm_120a/f and sm_121a/f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM accepted the final paired fixture with offset 16 (source SHA-256 110e6ae381ae3cb6e810814279a6bdccce57510a39f4b5489987ab58a30ed607). Direct PTX retained 32 entries, 16 loads, 16 stores, 8 pack::16b forms, and 8 unpack::16b forms. PTX SHA-256: compute_100a 0b69ee77a14d958c3857778d2bc6d95216a169793ec93d457c8f7d985cc2d59e; compute_100f 000a052d030990964ad52d3528ec0eaca1aad1fd74660ab7055a937bcf91198f; compute_103a 657fef7ae567c5e422be96428699472b09e399b06df72218e61deacad4320f01; compute_103f 5dc48412591c8c440ce988e129845c7569c1554ebf3c063f2ab41eb2df73e618; compute_110a bc86af044bea85df48423d1753a53a711a59bc569324cc97ca3f67987b155c4a; compute_110f 8814e93e62924a4926ed737a2e5f4d9db061895e27c289e2e04a80bafc12b7bc. LTOIR SHA-256: sm_100a ec97940a75fa160ed9b0ef9ff5dc582abbf536a7da9148d05d2f661c36ef8ac4; sm_100f ee252c6fd3c005fcb55113c77df0e477000055604e5a6fc042d80fd12e7df11a; sm_103a 69cd047c5845166a3da070b39403ac54ade7cc6485bd7512090a54952e5ec695; sm_103f c13ca599567deaa87eb4f777009e05ffaa284be16c5ae656e69296e8dabd17e9; sm_110a 47a504ee4d1048fdca4fa38c437ee8bac5534fee0a8905c37a9640a3a9e580df; sm_110f 88ac04e5505b1306b30fd365823631e1357a9713cd3d71d44fc5046c0fb47bbe. Tool `/usr/local/cuda-13.3/nvvm/lib64/libnvvm.so.4.0.0` reports `CUDA toolkit 13.3; libNVVM 2.0; NVVM IR 2.0; debug IR 3.2; release V13.3.33` (SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`).
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a and compute_101f with NVVM_ERROR_INVALID_OPTION because neither architecture spelling is supported.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced the same terminal cubins as the pinned-LLVM PTX route: sm_100a 31c10575f4abc1e08b67332565641d620d9174d72f3be907f654fea38fc3449d; sm_100f f222a38081bea23819ddf67f48c8e14b1672513d59e4c47fcad5c726ece892fc; sm_103a 558a6b8b226bc089613d31589e40ff9f4bbcb976f07d20d72c3b70d3767330c3; sm_103f 9766d55146e1be1d52ca6f47b0ee059c23aa6dbf1f8e3e908e7fd16bd0ed92e9; sm_110a ee853129023a8a296d0c5f46ff33562dcb97b8ee85aa4f5872d01bbf5c6a38ed; sm_110f d1f14616cd2fb70649cea20be2d42f11f7c613ffebcd1a295e44180d3b32828e. Each has 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink cubin SHA-256: sm_100a 81bfb82f0a8eed5ffbd37066c79b46e0c6b5302b2e293872fcbe5e5b929efd15; sm_100f e43b2afba55de137df03c68d998ead08d9dbd2052de984771a92d0c4421fbb9c; sm_103a b443ed7f8306f0cf48460dc4be4c067686463c1238ea73295fec70dea2111a33; sm_103f ff0489ac2481b5eaa2437cd9b009da8e7760806221a1823e7315b376438ad49c; sm_110a 1725799b8e2212f4fff9b7a6fde12778fc1c03dc0efbc91230ce1e9f8e96a9d2; sm_110f 6e8e5536d22abd70c8c91c0bbf1b65f9b9519311bfc51cee68c6ed202f24acb2. Every linked cubin retains 32 kernel sections, 32 LDTM, 32 STTM, 16 PACK16BIT, 16 EXPAND16BIT, 32 base halves, and 32 second halves at +0x10. The reviewed libNVVM policy may remain exact to sm_100a, sm_103a, and sm_110a; family-target results are separately proven evidence. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x128_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x128_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x16_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x1_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x2_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x32_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x4_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x64_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_16x64b_x8_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x128_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x128_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x16_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x16_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x1_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x1_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x2_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x2_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x32_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x32_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x4_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x4_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x64_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x64_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x8_raw`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_st_32x32b_x8_unpack16`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. Pinned LLVM 22 at -O0 accepted the final 58-kernel inline-PTX fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). Every target retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms. PTX SHA-256 was 61cdae4e568211ec62d3a53df95b2645704a63fe4c96ad46265938e7fbff27b8 for sm_100a at PTX 8.6, a87efd34bcf8fa73cd165d4246212d9d28a6d5795fba634c86b64f7a0d9fd92d for sm_101a at PTX 8.6, f831021d303379af1cf5ecac9dde5db21478143e7d931bc86f1666f6717504dc for sm_103a at PTX 8.8, and 0372ad17014f26da0b5b9f990452fcbdb665c662f84b0d19ac5db92a1f284d9a for sm_110a at PTX 9.0. The reviewed direct route is exact to those four architectures.
-    - backend codegen on `sm_100a, ptx85`: `failed` — Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. CUDA 13.3 ptxas has no sm_101a gpu-name spelling; its final-source PTX still contains all 58 entry kernels and exact stores. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
-    - PTX assembly on `sm_120a, ptx87`: `failed` — CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 80fed328663eb07e466dd82913bfc5c0a0649e559f5fb93e44ac51241abf7373.
+    - backend codegen on `sm_100a, ptx85`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. Pinned LLVM rejected sm_100a with PTX 8.5 before emission because this target requires PTX 8.6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 58 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5fa98e6dbf532a61fa907f4d0c05db8e62586087e077cd4530f3e59aaf160f27. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
+    - PTX assembly on `sm_120a, ptx87`: `not_revalidated` — Historical evidence recorded for Rust e457a7b0d326d67b4322ef0d11bd715cfaeda48f, LLVM 16696adcd119e6ba9cc175207d984d7021211acb (LLVM version 23.1.0-rust-1.100.0-nightly). Not revalidated by the 2026-10-11 refresh: Original outcome: failed. CUDA 13.3 ptxas rejected tcgen05.st, all four store shapes, and unpack::16b on sm_120a. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`).
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05-st` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — The 29 imported LLVM records each have arguments [tmem_ptr, their exact scalar or vector data carrier, i1], no result, ImmArg<arg2>, IntrArgMemOnly, IntrConvergent, and NoCapture<arg0>. Each raw leaf fixes arg2 to false; each unpack16 leaf fixes arg2 to true. CUDA 13.3 libNVVM accepted the final 58-kernel fixture (source SHA-256 2ba0edc2f7c5619537149035d8803599640c6f1c2555e83b50b4aca8d4922155). LTOIR SHA-256 was 1072cfb428e547f83140f1e3f51a5d4ca9f2a9f6115ae018bd600f2fca0ff207 for compute_100a, 2e75cdaa47d1ef47eb2b8772bb84a2a68b44a335b02ee2b910aaf69c87a0983e for compute_103a, and 117c4254b22d002b3507aa9194885ff779fd71badb845da42366d55f2c56e5ab for compute_110a. Direct PTX output retained 58 entry kernels, 58 exact stores, and 29 unpack::16b forms on every target; PTX SHA-256 was 407c66c108efdc7a61e1c371fa16fd736862a069b48c20d2f75adbf258e9f657 for sm_100a, 674dc740127d9e173e2f17119419b5e611c368dbc308fd7b2645048ba29b6ce9 for sm_103a, and 7a13a1fe0035b71504d568f03f753accdb0a0a142ff60fae6ba62844738749eb for sm_110a; each is PTX 9.3.
     - backend codegen on `compute_101a, ptx93`: `failed` — CUDA 13.3 libNVVM rejected compute_101a before code generation with NVVM_ERROR_INVALID_OPTION and reported that -arch=compute_101a is unsupported, so this backend route excludes sm_101a.
     - PTX assembly on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 ptxas produced cubin SHA-256 9d9144fc7f3e4f8cb6a330544f6186cbef958141fedc02d80726269159a70ccc for sm_100a, d13b796aee65324b290ac1a0c06964897ac35e09667f7ce26a1fdba3854d24d7 for sm_103a, and 4691bc7245af55f295b25c824be0b43d253230d29e4f17d523dc5449aa1c408a for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced cubin SHA-256 954aae70979f441ed4845a0e27609594431eb46f5a77f9424d10e559b58f87d7 for sm_100a, 750c2316839064b62bd30638739bbb977fbd17e7417a34ad0bec9a9771514ce4 for sm_103a, and 213abf3b4928f63629e1836e157da836979da9f75e7cf610e278f4545764dc43 for sm_110a. Each cubin contains 58 entry symbols and nvdisasm found exactly 58 STTM instructions, including 29 EXPAND16BIT forms. The reviewed libNVVM route is exact to sm_100a, sm_103a, and sm_110a. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tcgen05_store_wait`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — Pinned LLVM 22 accepted the 9,312-byte source (SHA-256 01af9baf38e4bf0cb8101163455b3e8255d98d4c7cf3488cc7adb9543038fc6e), produced 9,244-byte bitcode (SHA-256 77011a9c84fbc53e36b63f074b471025e94a5b9a47eadbaa20e4d58d541711aa), and retained all 24 exact instructions. PTX SHA-256 was 501f6c2d213c73d82415acba415756ff5c2c74efd52b6a0f7685e3d1eb8ec4e2 for sm_100a, 76a6707a7d232e6d3a3f62918d29e87996bf8d1f0ca5ed35ece6a4624921a8e8 for sm_101a, bdcfe7903077855ff7f40ce0b7a9902be054ebd3bc596789b4549336ea67bbae for sm_103a, and 7b0f6907d1db4f58adbb9d94f82a2dcf0d4bc5ff8e5cec918e51b635f56c737d for sm_110a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted all 24 routes on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was e36d4a8295fb2b51a11aba88b85fd5c2cab86842963f048969b6a99c219766a6, 3a290520b622b59caae3186c109b80d1a5e79e2c24a93fc5611225dfcd39b829, and 245611da00f4fda4886614f0d49f9bf1b1a548d5002f4a31b58ec54093de65f4. CUDA 13.3 has no sm_101a assembler spelling; LLVM 22 supplies the compatibility spelling. The same exact fixture was rejected for every family on sm_120a and sm_121a, so those architectures are deliberately outside this closed target contract. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 8.6 / sm_100a or sm_101a or sm_103a or sm_110a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 5200b0e376515627fd3b7fa095e31dd4460af0b45affd7110ea8da1063d9c0a6.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 24 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 3874345fd2dfbae0441a2afabcf498dd4eb3f009fc0dd0b4876672dc9607329f. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-tcgen05` at PTX 8.6 / sm_100a or sm_103a or sm_110a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_100a, ptx93`: `succeeded` — CUDA 13.3 libNVVM verified the combined module and retained all 24 exact tcgen05 instructions through LTOIR and linked PTX. LTOIR SHA-256 was 234ddb423fd986e4ac7df3c50c265734fe851f8fcd3fc97827a5a70f7c1f56b3 for compute_100a, ca5840112666fac59bde59b8a7777f66bc864e5a8e2832a122b458ce38485f50 for compute_103a, and 3545a605caea9c50008d7ecb8ab1c41f7ffeea129e791baf79eef522608f33bf for compute_110a. Linked PTX SHA-256 was 18b127dd78c7a7ed3ca0b08a1e24d73c0ca32e2d29d51c726ddbde71f60eb7a9, 7b8944889d6d4ee5ce99639ce869f841437ef9b181f4449eb402c9004844758b, and 06431b22f61484cf3b51e7e2090acd747f4a515a219cbf53bdec284797a1729c. A direct compute_101a probe failed before code generation with NVVM_ERROR_INVALID_INPUT, so this backend route excludes sm_101a. The BF16 public route intentionally emitted PTX kind::f16 and no kind::bf16.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink accepted every route on sm_100a, sm_103a, and sm_110a. Cubin SHA-256 was 1bab3647f9af53b36a52caa8323aeeacdf4244386f1ab0584e9823dcc1f13014, 57f3f660825a5d2970b34127c72cee000291ae352e4a33b508998bb0410d8a75, and 9bb5d9ded614b742960d19ed430d0f91a0fb8c90e675c7912c664cbad695a2de. The same exact LTO route reached PTX but cubin linking rejected every family on sm_120a and sm_121a; those architectures are deliberately excluded. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `tensormap_replace_box_dim`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_element_stride`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_element_type`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_fill_mode`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_global_address`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_global_dim`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_global_stride`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_interleave_layout`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_rank`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_swizzle_atomicity`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_100a, ptx86`: `succeeded` — The pinned LLVM retained the exact instruction in PTX; the ID/PTX-hash list has SHA-256 588486907e9e70101b6445f23f07d27798a009766a831e519c21a0a2fe4ea178.
-    - PTX assembly on `sm_100a, ptx86`: `succeeded` — CUDA 13.3 ptxas accepted the module; the ID/cubin-hash list has SHA-256 4bfa0ad94eac3615d96e64ef7594b7b7b96b549057da9dd83f38afd135c5aed1. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_100a, ptx86`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 02683e3a77ca0ae4f160645826b8e3da2b9da126376f94ce15265ad34e8ca282.
+    - PTX assembly on `sm_100a, ptx86`: `succeeded` — Revalidated 1 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9d36f1191de4ceed2adb6d065c793c92620591a4fbd3bafe95571a43268f4ec3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.6 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_100a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained the exact instruction. The source IR SHA-256 is 5c89ab7501c8bcf0af4dff26a5d3b0d3da1e5174955e14a54368c4898885394b, LTOIR SHA-256 is 567f2aacbdb7c5161f9a06121a97afc9c5bcff8091dc605b37646bd3d9cd0ce5, and linked PTX SHA-256 is f94d624b6cee55a630f4bb7fefcfcf4a71e6536332fcbe7decde2017e4c1973c.
     - device link on `sm_100a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a valid CUDA ELF cubin with SHA-256 45549b6566d359c474e2765940fb4143e182d8db51b7318317b076670a041b5a. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `tensormap_replace_swizzle_mode`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_90a, ptx83`: `succeeded` — The pinned LLVM retained every exact instruction; the sorted per-operation ID/PTX-hash list has SHA-256 f78503c587269a0cdfb58894b9848f8164fc545eff992e8db786c7af98f36552.
-    - PTX assembly on `sm_90a, ptx83`: `succeeded` — CUDA 13.3 ptxas accepted every module; the sorted per-operation ID/cubin-hash list has SHA-256 ed6ad097cb8cbda801025afd575d96b03f5b55eee07cd388b44908a1972fb589. Tool `ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73` (SHA-256 `ffb76e4665a12915797c67adf37153fc1a1253ec1ebc3a4d3cbe92a0112b310f`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 c13def3afba67facaae52854bb899bcac08cc9c1302bc94df9adb774203fb9dc.
+    - PTX assembly on `sm_90a, ptx83`: `succeeded` — Revalidated 10 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 a25c2e046ebef3f1a8fed295a64b6472c2dda0d7ee0a5cb1d6c04c83b3827391. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.73-tma` at PTX 8.3 / sm_100a or sm_100f or sm_103a or sm_103f or sm_110a or sm_110f or sm_120a or sm_120f or sm_121a or sm_121f or sm_90a: status `validated` (`CUDA 13.3 libNVVM package 13.3.73; libNVVM API 2.0`, SHA-256 `5317a9fb3c8b719198e89e27fd60e2b59f2702fe6b1eab357167c920b52be9cc`)
     - backend codegen on `sm_90a, ptx93`: `succeeded` — CUDA 13.3.73 libNVVM retained every exact instruction. SHA-256 over sorted per-operation ID/hash lists is e2fc2d11cb77ceac70ba08c26b914df1b0de4de38235f819941486ad1c9d9a73 for source IR, 0874909a45a8726ef5f8eb53835e57049511be19732706f6162ef60b49a08255 for LTOIR, and fa4efe247f38547bcd0ca3a85b988d12868499c426af9129c862963949bc5f04 for linked PTX.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced valid CUDA ELF cubins; the sorted per-operation ID/cubin-hash list has SHA-256 25167392d8f71be485b001b3f2d17c1d081188b8a4e8f0cb122a465d011046f9. Tool `libnvJitLink.so.13.3.33` reports `CUDA 13.3 nvJitLink 13.3.33` (SHA-256 `f84abd240b428825c3ef16632deb1026b208252fe19cef225a9f4ad06a39d1aa`). Artifact `cubin`.
 - `threadfence`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-threadfence` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_80`: `succeeded` — CUDA 13.3 libNVVM verified the 614-byte source module with SHA-256 f11112eb44a00edcb349bd511051277471941e522df16085960dfe4c3826b5f6.
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM emitted 319-byte PTX containing each exact instruction once, with SHA-256 d8fa0b45f2670cef336d07f5adb669d73c5a9789a89e029d5a2d0ecf5e4add16.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced PTX into a 3,496-byte cubin with SHA-256 32485fd3021ef42b89a78cd410e367961b4a050ef2e8bb455eebab9d8d182b87. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `threadfence_block`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-threadfence` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_80`: `succeeded` — CUDA 13.3 libNVVM verified the 614-byte source module with SHA-256 f11112eb44a00edcb349bd511051277471941e522df16085960dfe4c3826b5f6.
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM emitted 319-byte PTX containing each exact instruction once, with SHA-256 d8fa0b45f2670cef336d07f5adb669d73c5a9789a89e029d5a2d0ecf5e4add16.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced PTX into a 3,496-byte cubin with SHA-256 32485fd3021ef42b89a78cd410e367961b4a050ef2e8bb455eebab9d8d182b87. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `threadfence_system`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_20`: `succeeded` — The pinned llvm-as and llvm-dis accepted and canonicalized all three imported declarations. Canonical IR SHA-256 values were b1334ac980609185fb4e67a6327e6c5309e5b6748e9c2934a6a6fdb29d04a0de, 39231f5eea921f21c31c5df5e129677641609b2ee338a33ce8411807bdcec038, and b81cec3aa83bd01bc89549ffb35a36a1149d6aa99ae561fe4fbdd1432b727209.
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The pinned Rust LLVM emitted each exact instruction at the route floor. PTX SHA-256 values were 6524a3ea94ed3a1bfae679c04c4a9d98e85c9403392e7e9cb9eff2ae7804927a, 2467569ba7715d2ce6b6d18e3388a03afabe360610a6333107540b97c4ef24c5, and e95fece1b6217af0fcead2b754bbeaae60e8c1ee3ded17133a7be0e3dfc0e0f3.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted every module and produced 1,464-byte cubins with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_20`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 cdb1b77f91592ace02059db93a2d8c3f5e2698ede82912e1892a424538bde68a.
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 6e3c80aa9c5550fa1113e11c01c24c7e75778513305d9aea3bf1cd2efd6b391a.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 05ab6ee6476e257910ffd4fcc8e706687f44f9ea46da4a8adf146207e8feb62b. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `typed_nvvm` from profile `cuda-13.3-libnvvm-13.3.33-threadfence` at PTX 7.0 / sm_80+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - declaration canonicalization on `sm_80`: `succeeded` — CUDA 13.3 libNVVM verified the 614-byte source module with SHA-256 f11112eb44a00edcb349bd511051277471941e522df16085960dfe4c3826b5f6.
     - backend codegen on `sm_80, ptx93`: `succeeded` — CUDA 13.3 libNVVM emitted 319-byte PTX containing each exact instruction once, with SHA-256 d8fa0b45f2670cef336d07f5adb669d73c5a9789a89e029d5a2d0ecf5e4add16.
     - device link on `sm_80, ptx93`: `succeeded` — CUDA 13.3 nvJitLink linked the libNVVM-produced PTX into a 3,496-byte cubin with SHA-256 32485fd3021ef42b89a78cd410e367961b4a050ef2e8bb455eebab9d8d182b87. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `total_smem_size`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 4.1 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx41`: `succeeded` — Pinned llc emitted both size reads in 655-byte PTX with SHA-256 4ab2630ee33912457ed45c1cd516ae0797e062fcba7123db9bac352c67280b3a.
-    - PTX assembly on `sm_75, ptx41`: `succeeded` — CUDA 13.3 ptxas produced a 2,856-byte cubin with SHA-256 ee3358b59d2b596f1b276d983c3149d58c384872ee5edad482cf20a99183e7b1. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 4.1 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx41`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 24cc1707db68cd33a711956ac50c8c65ab3e84eb5cf71a00fc15160d770b645d.
+    - PTX assembly on `sm_75, ptx41`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 7feb7ce1442470cb7ca3f613338c9fa901e16b23bc621a84ac3aa46a8583088d. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx41, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 4.1 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained globaltimer, smid, gridid, warpid, dynamic_smem_size, and total_smem_size in 2,028-byte LTOIR (SHA-256 e00f26c9c773f5ea24198cda4a679a0de54ceafb3a13be5190b4dcf0b475f864) and emitted 1,135-byte PTX (SHA-256 4cb7846fcc9a10b3d26ccbab61c1936f26ae6255911d637cf6875fbf4e6d3bea).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 60456548b55c17dbef4e6a269c3eacb20298af44406ed81eec5657362501cdda; cuobjdump retained globaltimer, virtual SM, grid, warp, and shared-memory size reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `trap`: runtime `unexecuted`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc-debug-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx32`: `succeeded` — The pinned Rust LLVM emitted PTX 3.2 containing all three instructions; source SHA-256 cbe8cd68385d281ecb4edeb5458a63f1e851ee0ab2539c7b5d335b79e91e543e and PTX SHA-256 df263240edffc345f2c5a65b953c362a49878e08e9fdf1c2a06858829643dad5.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 1,464-byte cubin with SHA-256 c7cba4e51e90a77cda2ea1bf2ce9194bffd4a833d4e55f9909b91441d4cbaa91. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0-debug-control` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 f27ba289b79e2a2f272296bb6becce15edfbb0b95895792b50a304e0dc5e32b5.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 79791b5f0706b86706617c4633830f5c8540ccd30af7c03f507116cfececd645. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-debug-control` at PTX 3.2 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 emitted a 562-byte PTX 9.3 module containing all three instructions; source SHA-256 2e8b8ffd062afdcf1c1806d526388b9bacc7469918370d26af9186b2c9a4b1e1 and PTX SHA-256 ba2d42ec37ae06d4a460291a966872cbc35e999ffd0aa72c7e1aa6cf0f54e6f8.
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 produced a 5,416-byte cubin with SHA-256 d9c0af472531aa6da4a84efc0b4cbaccc56f9e090c5c3160792842172c225ee1; cuobjdump retained PMTRIG and BPT.TRAP instructions. Tool `/usr/local/cuda-13.3/bin/nvcc` reports `CUDA compilation tools, release 13.3, V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `6b55d91cd23f314ccb6801d71d64d20760a7ec31d37d87017ea0681d79bd3285`). Artifact `cubin`.
 - `warpid`: runtime `not recorded`
-  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.0-16696adc` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - backend codegen on `sm_20, ptx32`: `succeeded` — Pinned llc emitted both volatile location samples in 633-byte PTX with SHA-256 8b6bdea0bf90c68baecd7189ae84ba5c24a0ca32a3a347a1dc72ca8a0a62277f.
-    - PTX assembly on `sm_75, ptx32`: `succeeded` — CUDA 13.3 ptxas produced a 2,984-byte cubin with SHA-256 2a1f9febb72d01384c5c1c4583e4511020b2c086144d8b25f6aad354a4c34618. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `inline_ptx` from profile `rust-llvm-23.1.3-aab11ee0` at PTX 3.2 / sm_20+: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - backend codegen on `sm_20, ptx32`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 9ec007869a74a8d4d9bf7be114f276c2c4e26f3e4c34eeaab721362bbf9a1918.
+    - PTX assembly on `sm_75, ptx32`: `succeeded` — Revalidated 2 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 14f47ee4cc2165c26a56b268fb647e325de01ae94e284257de734b97dea0626a. LLVM codegen and terminal assembly use distinct recorded targets: llc sm_20 +ptx32, ptxas -arch=sm_75. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33` at PTX 1.3 / sm_75+: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `sm_75, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained globaltimer, smid, gridid, warpid, dynamic_smem_size, and total_smem_size in 2,028-byte LTOIR (SHA-256 e00f26c9c773f5ea24198cda4a679a0de54ceafb3a13be5190b4dcf0b475f864) and emitted 1,135-byte PTX (SHA-256 4cb7846fcc9a10b3d26ccbab61c1936f26ae6255911d637cf6875fbf4e6d3bea).
     - device link on `sm_75, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,240-byte cubin with SHA-256 60456548b55c17dbef4e6a269c3eacb20298af44406ed81eec5657362501cdda; cuobjdump retained globaltimer, virtual SM, grid, warp, and shared-memory size reads. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `wgmma_commit_group`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90a`: `succeeded` — llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f.
-    - backend codegen on `sm_90a, ptx80`: `succeeded` — The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98.
-    - PTX assembly on `sm_90a, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90a`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8.
+    - backend codegen on `sm_90a, ptx80`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3.
+    - PTX assembly on `sm_90a, ptx80`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90a, ptx93`: `failed` — libNVVM accepted the 864-byte source with SHA-256 9688c2fd5593eae29ec346f031fe10414406bbad1035ecd83c60f7106900a480 but emitted three unresolved extern calls. The 2,052-byte LTOIR had SHA-256 cd7b151407b8838020b4be909b2736482404de9477864c9a8a2ac9251685e293; the 937-byte PTX had SHA-256 2470f349c91be972eda200df903bc4092a5d4ebad89d08e31f231b2f36c41a43.
     - device link on `sm_90a, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after ptxas rejected the unresolved .extern names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `compute_90a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained each exact instruction once. The 760-byte source had SHA-256 cbd3c0a7fbaf951487fd7edf28eb9a841499ad5c277ab27b3eb03817f70f9d3c; the 2,040-byte LTOIR had SHA-256 a648e9385b3863bc44368471bf2a710c22e94ec68af423466b77d4fbb88d1037; the 491-byte linked PTX had SHA-256 5423249b889c249c36732c46893c3987fef414e060f48852a29887bd918fad71.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,560-byte cubin with SHA-256 a3ebc2a2c4d8a7661a85f5e06421cd24b7f35a37754509b651be7ed1c0783ae5. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `wgmma_fence`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90a`: `succeeded` — llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f.
-    - backend codegen on `sm_90a, ptx80`: `succeeded` — The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98.
-    - PTX assembly on `sm_90a, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90a`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8.
+    - backend codegen on `sm_90a, ptx80`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3.
+    - PTX assembly on `sm_90a, ptx80`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90a, ptx93`: `failed` — libNVVM accepted the 864-byte source with SHA-256 9688c2fd5593eae29ec346f031fe10414406bbad1035ecd83c60f7106900a480 but emitted three unresolved extern calls. The 2,052-byte LTOIR had SHA-256 cd7b151407b8838020b4be909b2736482404de9477864c9a8a2ac9251685e293; the 937-byte PTX had SHA-256 2470f349c91be972eda200df903bc4092a5d4ebad89d08e31f231b2f36c41a43.
     - device link on `sm_90a, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after ptxas rejected the unresolved .extern names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
     - backend codegen on `compute_90a, ptx93`: `succeeded` — CUDA 13.3 libNVVM retained each exact instruction once. The 760-byte source had SHA-256 cbd3c0a7fbaf951487fd7edf28eb9a841499ad5c277ab27b3eb03817f70f9d3c; the 2,040-byte LTOIR had SHA-256 a648e9385b3863bc44368471bf2a710c22e94ec68af423466b77d4fbb88d1037; the 491-byte linked PTX had SHA-256 5423249b889c249c36732c46893c3987fef414e060f48852a29887bd918fad71.
     - device link on `sm_90a, ptx93`: `succeeded` — CUDA 13.3 nvJitLink produced a 3,560-byte cubin with SHA-256 a3ebc2a2c4d8a7661a85f5e06421cd24b7f35a37754509b651be7ed1c0783ae5. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`). Artifact `cubin`.
 - `wgmma_wait_group`: runtime `not recorded`
-  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.0-16696adc-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`LLVM version 23.1.0-rust-1.100.0-nightly`, SHA-256 `2e5bc688625ca7d98bca9b5f0f0d14525d2eb59e8a9842964e480df3ffd1f921`)
-    - declaration canonicalization on `sm_90a`: `succeeded` — llvm-as accepted the convergent declarations and wait-group immarg. The 561-byte source had SHA-256 c1bff19ab6f92fda30c1c832f2c2322ab2fc66c2615454e653d4b8e6be1e8172; the 2,132-byte bitcode had SHA-256 67444d4b8d1b6bdcc67db2375ce0dd87545aaec2630e611d6a38c1c0f8c4921f.
-    - backend codegen on `sm_90a, ptx80`: `succeeded` — The pinned Rust LLVM emitted each exact instruction once. The 484-byte PTX had SHA-256 e634738c624de6c258c4988201d73b5f027f3eaf8e64ee5034288d8bd2802d98.
-    - PTX assembly on `sm_90a, ptx80`: `succeeded` — CUDA 13.3 ptxas accepted the module and produced a 3,560-byte cubin with SHA-256 38253065117f4e13a468c91f3dc2c55f375b47fad59da34892623c49604283c3. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `CUDA 13.3 V13.3.33 build cuda_13.3.r13.3/compiler.37862127_0` (SHA-256 `7fdd01a4cf50e30746da98989c9272a907f491e6fd7fecfda14642e4375f88fb`). Artifact `cubin`.
+  - `llvm_nvptx` uses `typed_nvvm` from profile `rust-llvm-23.1.3-aab11ee0-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`LLVM version 23.1.3-rust-1.101.0-nightly`, SHA-256 `7534b0236b9b26c7e6910548038b2285ee1bb5908e4700764dc78a7a1a54b449`)
+    - declaration canonicalization on `sm_90a`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): canonical typed declarations and their exact imported attributes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ab03f3000111d656cd57565aab9548160995c8d8c9ca47c095281df73aa8c4d8.
+    - backend codegen on `sm_90a, ptx80`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): the exact reviewed PTX instruction shapes passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 ecbe002187ec15686ef6eeb8f3a9bed78aaecd272bc22efadeeb869a6f834cd3.
+    - PTX assembly on `sm_90a, ptx80`: `succeeded` — Revalidated 3 separate probes with nightly-2026-10-11 (rustc 6f5da406b56c2db6607202b63daee73aa8c61297; LLVM aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019): terminal PTX assembly passed. The sorted ID/artifact-kind/SHA-256 ledger in refresh-2026-10-11/stages.jsonl has SHA-256 2fa706ef170664422171110b7e74547615f5bbc7438a0997cb5a5c9aaaab4af8. Tool `/usr/local/cuda-13.3/bin/ptxas` reports `Cuda compilation tools, release 13.3, V13.3.73; build cuda_13.3.r13.3/compiler.38244171_0` (SHA-256 `afd8d1e1fa6e310f7faee44f6621e4c1315fb7fd6da7d4d87414358e12a651dc`). Artifact `cubin`.
   - `lib_nvvm` uses `inline_ptx` from profile `cuda-13.3-libnvvm-13.3.33-wgmma-control` at PTX 8.0 / sm_90a: status `validated` (`CUDA toolkit 13.3; nvcc/libNVVM release V13.3.33`, SHA-256 `b6a5db233133fb0ed693f8cb8aeb1d405254ce0ab10364478985292ec98cfcf0`)
     - backend codegen on `compute_90a, ptx93`: `failed` — libNVVM accepted the 864-byte source with SHA-256 9688c2fd5593eae29ec346f031fe10414406bbad1035ecd83c60f7106900a480 but emitted three unresolved extern calls. The 2,052-byte LTOIR had SHA-256 cd7b151407b8838020b4be909b2736482404de9477864c9a8a2ac9251685e293; the 937-byte PTX had SHA-256 2470f349c91be972eda200df903bc4092a5d4ebad89d08e31f231b2f36c41a43.
     - device link on `sm_90a, ptx93`: `failed` — CUDA 13.3 nvJitLink exited with code 22 after ptxas rejected the unresolved .extern names; no cubin was produced. Tool `/usr/local/cuda-13.3/lib64/libnvJitLink.so.13.3.33` reports `V13.3.33` (SHA-256 `3ba1e744347cd68617b862eccfd98b125482e882b7a6319f42abc9a768513db8`).
diff --git a/cuda-oxide/intrinsics/imported.json b/cuda-oxide/intrinsics/imported.json
index 9a00dde104..f47e97912a 100644
--- a/cuda-oxide/intrinsics/imported.json
+++ b/cuda-oxide/intrinsics/imported.json
@@ -2,9 +2,9 @@
   "schema": 2,
   "source": {
     "llvm_repository": "https://github.com/rust-lang/llvm-project.git",
-    "llvm_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
-    "llvm_tblgen_version": "LLVM version 23.1.0",
-    "llvm_tblgen_source_revision": "16696adcd119e6ba9cc175207d984d7021211acb",
+    "llvm_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
+    "llvm_tblgen_version": "LLVM version 23.1.3",
+    "llvm_tblgen_source_revision": "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019",
     "intrinsics_json_sha256": "c0fb05eda32f2fe89ba600d4d2dc882e0bf54d5aa60b5e5f8601eb87a9881408",
     "nvptx_json_sha256": "492011cd3c20fbd888a2fa05e7eb5d894c0db79270c5bf6dbda0b6648d3c6582"
   },
diff --git a/cuda-oxide/intrinsics/overlay.toml b/cuda-oxide/intrinsics/overlay.toml
index 1a4e179e5e..1ba76db28d 100644
--- a/cuda-oxide/intrinsics/overlay.toml
+++ b/cuda-oxide/intrinsics/overlay.toml
@@ -1,7 +1,7 @@
 schema = 44
 catalog_version = "0.1.0"
 intrinsic_abi = 1
-backend_profile = "rust-llvm-23.1.0-16696adc"
+backend_profile = "rust-llvm-23.1.3-aab11ee0"
 shards = [
     "overlay/active_mask.toml",
     "overlay/clc.toml",
diff --git a/cuda-oxide/intrinsics/overlay/active_mask.toml b/cuda-oxide/intrinsics/overlay/active_mask.toml
index ea57801b57..d01616f3da 100644
--- a/cuda-oxide/intrinsics/overlay/active_mask.toml
+++ b/cuda-oxide/intrinsics/overlay/active_mask.toml
@@ -38,7 +38,7 @@ ptx_isa_section = "9.7.14.12 Parallel Synchronization and Communication Instruct
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-activemask"
 lowering = "generated_active_mask"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.2", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.2", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_sm = "sm_75" },
 ]
 active_mask = { observation = "executing_lanes_at_instruction", adapter = "direct_zero_operand_mask" }
diff --git a/cuda-oxide/intrinsics/overlay/clc.toml b/cuda-oxide/intrinsics/overlay/clc.toml
index d5b738203e..f6776ac54f 100644
--- a/cuda-oxide/intrinsics/overlay/clc.toml
+++ b/cuda-oxide/intrinsics/overlay/clc.toml
@@ -4,7 +4,7 @@ schema = 40
 family = "clc"
 
 [clc]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-clc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-clc"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-clc"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/cluster_barrier.toml b/cuda-oxide/intrinsics/overlay/cluster_barrier.toml
index d206f33586..fdeb7033a4 100644
--- a/cuda-oxide/intrinsics/overlay/cluster_barrier.toml
+++ b/cuda-oxide/intrinsics/overlay/cluster_barrier.toml
@@ -4,7 +4,7 @@ schema = 31
 family = "cluster_barrier"
 
 [cluster_barrier]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-cluster-barrier"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/cluster_memory.toml b/cuda-oxide/intrinsics/overlay/cluster_memory.toml
index 992e813d5f..f5008d56d2 100644
--- a/cuda-oxide/intrinsics/overlay/cluster_memory.toml
+++ b/cuda-oxide/intrinsics/overlay/cluster_memory.toml
@@ -4,7 +4,7 @@ schema = 39
 family = "cluster_memory"
 
 [cluster_memory]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-cluster-memory"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-cluster-memory"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-cluster-memory"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/counted_barrier.toml b/cuda-oxide/intrinsics/overlay/counted_barrier.toml
index 410745460f..3bb09642f3 100644
--- a/cuda-oxide/intrinsics/overlay/counted_barrier.toml
+++ b/cuda-oxide/intrinsics/overlay/counted_barrier.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.14.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 expected_ptx = { mnemonic = "barrier", modifiers = ["sync"], operands = [{ kind = "register_or_immediate" }, { kind = "register_or_immediate" }] }
@@ -77,7 +77,7 @@ ptx_isa_section = "9.7.14.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "3.2", minimum_sm = "sm_20" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "3.2", minimum_sm = "sm_20" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "1.0", minimum_sm = "sm_75" },
 ]
 expected_ptx = { mnemonic = "bar", modifiers = ["sync"], operands = [{ kind = "register_or_immediate" }, { kind = "register_or_immediate" }] }
@@ -117,7 +117,7 @@ ptx_isa_section = "9.7.14.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 expected_ptx = { mnemonic = "barrier", modifiers = ["arrive"], operands = [{ kind = "register_or_immediate" }, { kind = "register_or_immediate" }] }
@@ -157,7 +157,7 @@ ptx_isa_section = "9.7.14.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "3.2", minimum_sm = "sm_20" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "3.2", minimum_sm = "sm_20" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "1.0", minimum_sm = "sm_75" },
 ]
 expected_ptx = { mnemonic = "bar", modifiers = ["arrive"], operands = [{ kind = "register_or_immediate" }, { kind = "register_or_immediate" }] }
diff --git a/cuda-oxide/intrinsics/overlay/cp_async_control.toml b/cuda-oxide/intrinsics/overlay/cp_async_control.toml
index 39d185926e..acfe9adf6e 100644
--- a/cuda-oxide/intrinsics/overlay/cp_async_control.toml
+++ b/cuda-oxide/intrinsics/overlay/cp_async_control.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.9.26.3.2 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-commit-group"
 lowering = "generated_cp_async_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_control = { operation = "commit_group", adapter = "no_operands", runtime_validation = "unexecuted" }
@@ -78,7 +78,7 @@ ptx_isa_section = "9.7.9.26.3.3 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all"
 lowering = "generated_cp_async_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_control = { operation = "wait_all", adapter = "no_operands", runtime_validation = "unexecuted" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.9.26.3.3 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async-wait-group-cp-async-wait-all"
 lowering = "generated_cp_async_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_control = { operation = "wait_group", adapter = "compile_time_constant_max_pending", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/cp_async_copy.toml b/cuda-oxide/intrinsics/overlay/cp_async_copy.toml
index 801b8a2dc1..fa6013e184 100644
--- a/cuda-oxide/intrinsics/overlay/cp_async_copy.toml
+++ b/cuda-oxide/intrinsics/overlay/cp_async_copy.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "ca", copy_size = "b4", source_size = "full", adapter = "direct_pointers", runtime_validation = "unexecuted" }
@@ -78,7 +78,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "ca", copy_size = "b4", source_size = "runtime", adapter = "direct_pointers_and_source_size", runtime_validation = "unexecuted" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "ca", copy_size = "b8", source_size = "full", adapter = "direct_pointers", runtime_validation = "unexecuted" }
@@ -160,7 +160,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "ca", copy_size = "b8", source_size = "runtime", adapter = "direct_pointers_and_source_size", runtime_validation = "unexecuted" }
@@ -201,7 +201,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "ca", copy_size = "b16", source_size = "full", adapter = "direct_pointers", runtime_validation = "unexecuted" }
@@ -242,7 +242,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "ca", copy_size = "b16", source_size = "runtime", adapter = "direct_pointers_and_source_size", runtime_validation = "unexecuted" }
@@ -283,7 +283,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "cg", copy_size = "b16", source_size = "full", adapter = "direct_pointers", runtime_validation = "unexecuted" }
@@ -324,7 +324,7 @@ ptx_isa_section = "9.7.9.26.3.1 Data Movement and Conversion Instructions: cp.as
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cp-async"
 lowering = "generated_cp_async_copy"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_copy = { cache_policy = "cg", copy_size = "b16", source_size = "runtime", adapter = "direct_pointers_and_source_size", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/cp_async_mbarrier.toml b/cuda-oxide/intrinsics/overlay/cp_async_mbarrier.toml
index 2d6552765a..db869543db 100644
--- a/cuda-oxide/intrinsics/overlay/cp_async_mbarrier.toml
+++ b/cuda-oxide/intrinsics/overlay/cp_async_mbarrier.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.14.16.18 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
 lowering = "generated_cp_async_mbarrier"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_mbarrier = { operation = "arrive", state_space = "generic", adapter = "pointer_to_void", runtime_validation = "unexecuted" }
@@ -78,7 +78,7 @@ ptx_isa_section = "9.7.14.16.18 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
 lowering = "generated_cp_async_mbarrier"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_mbarrier = { operation = "arrive", state_space = "shared", adapter = "pointer_to_void", runtime_validation = "unexecuted" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.14.16.18 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
 lowering = "generated_cp_async_mbarrier"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_mbarrier = { operation = "arrive_no_inc", state_space = "generic", adapter = "pointer_to_void", runtime_validation = "unexecuted" }
@@ -160,7 +160,7 @@ ptx_isa_section = "9.7.14.16.18 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-cp-async-mbarrier-arrive"
 lowering = "generated_cp_async_mbarrier"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 cp_async_mbarrier = { operation = "arrive_no_inc", state_space = "shared", adapter = "pointer_to_void", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/debug_control.toml b/cuda-oxide/intrinsics/overlay/debug_control.toml
index 8105fb77ac..ee380fc44d 100644
--- a/cuda-oxide/intrinsics/overlay/debug_control.toml
+++ b/cuda-oxide/intrinsics/overlay/debug_control.toml
@@ -2,7 +2,7 @@ schema = 33
 family = "debug_control"
 
 [debug_control]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-debug-control"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-debug-control"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-debug-control"
 runtime_validation = "unexecuted"
 operations = ["trap", "breakpoint", "pmevent"]
diff --git a/cuda-oxide/intrinsics/overlay/dotprod.toml b/cuda-oxide/intrinsics/overlay/dotprod.toml
index cf7201bade..0ba663cf32 100644
--- a/cuda-oxide/intrinsics/overlay/dotprod.toml
+++ b/cuda-oxide/intrinsics/overlay/dotprod.toml
@@ -38,7 +38,7 @@ ptx_isa_section = "9.7.1.24 Integer Arithmetic Instructions: dp4a"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a"
 lowering = "generated_dotprod"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_sm = "sm_75" },
 ]
 dot_product = { operation = "dp4a", signedness = "signed", adapter = "direct_three_operands" }
@@ -80,7 +80,7 @@ ptx_isa_section = "9.7.1.24 Integer Arithmetic Instructions: dp4a"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp4a"
 lowering = "generated_dotprod"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_sm = "sm_75" },
 ]
 dot_product = { operation = "dp4a", signedness = "unsigned", adapter = "direct_three_operands" }
@@ -122,7 +122,7 @@ ptx_isa_section = "9.7.1.25 Integer Arithmetic Instructions: dp2a"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a"
 lowering = "generated_dotprod"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_sm = "sm_75" },
 ]
 dot_product = { operation = "dp2a", signedness = "signed", adapter = "insert_low_half_false" }
@@ -164,7 +164,7 @@ ptx_isa_section = "9.7.1.25 Integer Arithmetic Instructions: dp2a"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-dp2a"
 lowering = "generated_dotprod"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_sm = "sm_75" },
 ]
 dot_product = { operation = "dp2a", signedness = "unsigned", adapter = "insert_low_half_false" }
diff --git a/cuda-oxide/intrinsics/overlay/elect.toml b/cuda-oxide/intrinsics/overlay/elect.toml
index 6cbefddd6e..b4d7d95347 100644
--- a/cuda-oxide/intrinsics/overlay/elect.toml
+++ b/cuda-oxide/intrinsics/overlay/elect.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.14.15 Parallel Synchronization and Communication Instruct
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-elect-sync"
 lowering = "generated_elect"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-elect", minimum_ptx = "8.0", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-elect", minimum_ptx = "8.0", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-elect", minimum_ptx = "8.0", minimum_sm = "sm_90" },
 ]
 expected_ptx = { mnemonic = "elect", modifiers = ["sync"], operands = [{ kind = "register_predicate_pair" }, { kind = "register_or_immediate" }] }
diff --git a/cuda-oxide/intrinsics/overlay/extended_minmax.toml b/cuda-oxide/intrinsics/overlay/extended_minmax.toml
index 847397cef5..cae206bcb8 100644
--- a/cuda-oxide/intrinsics/overlay/extended_minmax.toml
+++ b/cuda-oxide/intrinsics/overlay/extended_minmax.toml
@@ -4,7 +4,7 @@ schema = 60
 family = "extended_minmax"
 
 [extended_minmax]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-extended-minmax"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/grid_dependency.toml b/cuda-oxide/intrinsics/overlay/grid_dependency.toml
index 92c2534f1b..70363a9294 100644
--- a/cuda-oxide/intrinsics/overlay/grid_dependency.toml
+++ b/cuda-oxide/intrinsics/overlay/grid_dependency.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.15.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "7.8", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "7.8", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "7.8", minimum_sm = "sm_90" },
 ]
 expected_ptx = { mnemonic = "griddepcontrol", modifiers = ["launch_dependents"], operands = [] }
@@ -77,7 +77,7 @@ ptx_isa_section = "9.7.15.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-griddepcontrol"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "7.8", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "7.8", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "7.8", minimum_sm = "sm_90" },
 ]
 expected_ptx = { mnemonic = "griddepcontrol", modifiers = ["wait"], operands = [] }
diff --git a/cuda-oxide/intrinsics/overlay/integer_minmax.toml b/cuda-oxide/intrinsics/overlay/integer_minmax.toml
index bb2c105a88..4ffc0df52e 100644
--- a/cuda-oxide/intrinsics/overlay/integer_minmax.toml
+++ b/cuda-oxide/intrinsics/overlay/integer_minmax.toml
@@ -44,7 +44,7 @@ ptx_isa_section = "9.7.1.13 Integer Arithmetic Instructions: min"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "s32", native_minimum_sm = 90, operation = "min", relu = true }
 expected_ptx = { mnemonic = "min", modifiers = ["relu", "s32"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -82,7 +82,7 @@ ptx_isa_section = "9.7.1.14 Integer Arithmetic Instructions: max"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "s32", native_minimum_sm = 90, operation = "max", relu = true }
 expected_ptx = { mnemonic = "max", modifiers = ["relu", "s32"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -120,7 +120,7 @@ ptx_isa_section = "9.7.1.13 Integer Arithmetic Instructions: min"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "s16x2", native_minimum_sm = 90, operation = "min", relu = false }
 expected_ptx = { mnemonic = "min", modifiers = ["s16x2"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -158,7 +158,7 @@ ptx_isa_section = "9.7.1.14 Integer Arithmetic Instructions: max"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "s16x2", native_minimum_sm = 90, operation = "max", relu = false }
 expected_ptx = { mnemonic = "max", modifiers = ["s16x2"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -196,7 +196,7 @@ ptx_isa_section = "9.7.1.13 Integer Arithmetic Instructions: min"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "u16x2", native_minimum_sm = 90, operation = "min", relu = false }
 expected_ptx = { mnemonic = "min", modifiers = ["u16x2"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -234,7 +234,7 @@ ptx_isa_section = "9.7.1.14 Integer Arithmetic Instructions: max"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "u16x2", native_minimum_sm = 90, operation = "max", relu = false }
 expected_ptx = { mnemonic = "max", modifiers = ["u16x2"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -272,7 +272,7 @@ ptx_isa_section = "9.7.1.13 Integer Arithmetic Instructions: min"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-min"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "s16x2", native_minimum_sm = 90, operation = "min", relu = true }
 expected_ptx = { mnemonic = "min", modifiers = ["relu", "s16x2"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
@@ -310,7 +310,7 @@ ptx_isa_section = "9.7.1.14 Integer Arithmetic Instructions: max"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#integer-arithmetic-instructions-max"
 lowering = "generated_integer_minmax_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
 ]
 integer_minmax = { format = "s16x2", native_minimum_sm = 90, operation = "max", relu = true }
 expected_ptx = { mnemonic = "max", modifiers = ["relu", "s16x2"], operands = [{ kind = "register" }, { kind = "register" }, { kind = "register" }] }
diff --git a/cuda-oxide/intrinsics/overlay/ldmatrix.toml b/cuda-oxide/intrinsics/overlay/ldmatrix.toml
index 7852161da9..2019a1ea72 100644
--- a/cuda-oxide/intrinsics/overlay/ldmatrix.toml
+++ b/cuda-oxide/intrinsics/overlay/ldmatrix.toml
@@ -35,7 +35,7 @@ ptx_isa_section = "9.7.14.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 ldmatrix_variant = { shape = "m8n8", multiplicity = "x1", layout = "normal", element = "b16", state_space = "shared" }
@@ -77,7 +77,7 @@ ptx_isa_section = "9.7.14.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 ldmatrix_variant = { shape = "m8n8", multiplicity = "x1", layout = "transposed", element = "b16", state_space = "shared" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.14.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 ldmatrix_variant = { shape = "m8n8", multiplicity = "x2", layout = "normal", element = "b16", state_space = "shared" }
@@ -161,7 +161,7 @@ ptx_isa_section = "9.7.14.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 ldmatrix_variant = { shape = "m8n8", multiplicity = "x2", layout = "transposed", element = "b16", state_space = "shared" }
@@ -203,7 +203,7 @@ ptx_isa_section = "9.7.14.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 ldmatrix_variant = { shape = "m8n8", multiplicity = "x4", layout = "normal", element = "b16", state_space = "shared" }
@@ -245,7 +245,7 @@ ptx_isa_section = "9.7.14.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 ldmatrix_variant = { shape = "m8n8", multiplicity = "x4", layout = "transposed", element = "b16", state_space = "shared" }
diff --git a/cuda-oxide/intrinsics/overlay/ldmatrix_blackwell.toml b/cuda-oxide/intrinsics/overlay/ldmatrix_blackwell.toml
index 39f1cb3716..9b0f241c02 100644
--- a/cuda-oxide/intrinsics/overlay/ldmatrix_blackwell.toml
+++ b/cuda-oxide/intrinsics/overlay/ldmatrix_blackwell.toml
@@ -34,7 +34,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m16n16", multiplicity = "x1", layout = "transposed", element = "b8", state_space = "shared" }
@@ -75,7 +75,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m16n16", multiplicity = "x1", layout = "transposed", element = "b8x16_b4x16_p64", state_space = "shared" }
@@ -116,7 +116,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m16n16", multiplicity = "x1", layout = "transposed", element = "b8x16_b6x16_p32", state_space = "shared" }
@@ -157,7 +157,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m16n16", multiplicity = "x2", layout = "transposed", element = "b8", state_space = "shared" }
@@ -198,7 +198,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m16n16", multiplicity = "x2", layout = "transposed", element = "b8x16_b4x16_p64", state_space = "shared" }
@@ -239,7 +239,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m16n16", multiplicity = "x2", layout = "transposed", element = "b8x16_b6x16_p32", state_space = "shared" }
@@ -280,7 +280,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m8n16", multiplicity = "x1", layout = "normal", element = "b8x16_b4x16_p64", state_space = "shared" }
@@ -321,7 +321,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m8n16", multiplicity = "x1", layout = "normal", element = "b8x16_b6x16_p32", state_space = "shared" }
@@ -365,7 +365,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m8n16", multiplicity = "x2", layout = "normal", element = "b8x16_b4x16_p64", state_space = "shared" }
@@ -406,7 +406,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m8n16", multiplicity = "x2", layout = "normal", element = "b8x16_b6x16_p32", state_space = "shared" }
@@ -450,7 +450,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m8n16", multiplicity = "x4", layout = "normal", element = "b8x16_b4x16_p64", state_space = "shared" }
@@ -491,7 +491,7 @@ ptx_isa_section = "9.7.15.5.15 Warp-level matrix load instruction: ldmatrix"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-ldmatrix"
 lowering = "generated_ldmatrix"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-blackwell-ldmatrix" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-blackwell-ldmatrix" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-blackwell-ldmatrix", targets = "sm_100a|sm_100f|sm_103a|sm_103f|sm_110a|sm_110f|sm_120a|sm_120f|sm_121a|sm_121f" },
 ]
 ldmatrix_variant = { shape = "m8n16", multiplicity = "x4", layout = "normal", element = "b8x16_b6x16_p32", state_space = "shared" }
diff --git a/cuda-oxide/intrinsics/overlay/mbarrier_basic.toml b/cuda-oxide/intrinsics/overlay/mbarrier_basic.toml
index 88ec42a054..11c47a1d20 100644
--- a/cuda-oxide/intrinsics/overlay/mbarrier_basic.toml
+++ b/cuda-oxide/intrinsics/overlay/mbarrier_basic.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.14.16.12 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-init"
 lowering = "generated_mbarrier_basic"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 mbarrier_basic = { operation = "init", state_space = "shared", adapter = "pointer_count_to_void", runtime_validation = "unexecuted" }
@@ -78,7 +78,7 @@ ptx_isa_section = "9.7.14.16.16 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
 lowering = "generated_mbarrier_basic"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 mbarrier_basic = { operation = "arrive", state_space = "shared", adapter = "pointer_to_token", runtime_validation = "unexecuted" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.14.16.16 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-arrive"
 lowering = "generated_mbarrier_basic"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 mbarrier_basic = { operation = "arrive_no_complete", state_space = "shared", adapter = "pointer_count_to_token", runtime_validation = "unexecuted" }
@@ -160,7 +160,7 @@ ptx_isa_section = "9.7.14.16.19 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-test-wait-mbarrier-try-wait"
 lowering = "generated_mbarrier_basic"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 mbarrier_basic = { operation = "test_wait", state_space = "shared", adapter = "pointer_token_to_predicate", runtime_validation = "unexecuted" }
@@ -201,7 +201,7 @@ ptx_isa_section = "9.7.14.16.13 Parallel Synchronization and Communication Instr
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-mbarrier-inval"
 lowering = "generated_mbarrier_basic"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33" },
 ]
 mbarrier_basic = { operation = "inval", state_space = "shared", adapter = "pointer_to_void", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/mbarrier_extended.toml b/cuda-oxide/intrinsics/overlay/mbarrier_extended.toml
index a9d3794c91..a314be9244 100644
--- a/cuda-oxide/intrinsics/overlay/mbarrier_extended.toml
+++ b/cuda-oxide/intrinsics/overlay/mbarrier_extended.toml
@@ -4,7 +4,7 @@ schema = 40
 family = "mbarrier_extended"
 
 [mbarrier_extended]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-mbarrier-extended"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/movmatrix.toml b/cuda-oxide/intrinsics/overlay/movmatrix.toml
index e0d906850c..f50dac9d36 100644
--- a/cuda-oxide/intrinsics/overlay/movmatrix.toml
+++ b/cuda-oxide/intrinsics/overlay/movmatrix.toml
@@ -39,7 +39,7 @@ ptx_isa_section = "9.7.15.5.17 Warp-level matrix transpose instruction: movmatri
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-movmatrix"
 lowering = "generated_movmatrix_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.8", minimum_sm = "sm_75" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.8", minimum_sm = "sm_75" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33-movmatrix", minimum_ptx = "7.8", minimum_sm = "sm_75" },
 ]
 movmatrix = { participation = "all_warp_lanes_same_instruction_no_exited_lanes", adapter = "packed_b16x2_u32_to_packed_b16x2_u32", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/packed_alu.toml b/cuda-oxide/intrinsics/overlay/packed_alu.toml
index 804295fdad..696953ac7e 100644
--- a/cuda-oxide/intrinsics/overlay/packed_alu.toml
+++ b/cuda-oxide/intrinsics/overlay/packed_alu.toml
@@ -38,7 +38,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 80, operation = "fma", adapter = "direct_packed_u32" }
@@ -80,7 +80,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 80, operation = "fma_relu", adapter = "direct_packed_u32" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.4.1 Half Precision Floating Point Instructions: add"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.8", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.8", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.8", minimum_sm = "sm_90" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 90, operation = "add", adapter = "direct_packed_u32" }
@@ -158,7 +158,7 @@ ptx_isa_section = "9.7.4.2 Half Precision Floating Point Instructions: sub"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.8", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.8", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.8", minimum_sm = "sm_90" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 90, operation = "sub", adapter = "direct_packed_u32" }
@@ -197,7 +197,7 @@ ptx_isa_section = "9.7.4.3 Half Precision Floating Point Instructions: mul"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.8", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.8", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.8", minimum_sm = "sm_90" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 90, operation = "mul", adapter = "direct_packed_u32" }
@@ -239,7 +239,7 @@ ptx_isa_section = "9.7.4.7 Half Precision Floating Point Instructions: min"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 80, operation = "min", adapter = "direct_packed_u32" }
@@ -281,7 +281,7 @@ ptx_isa_section = "9.7.4.8 Half Precision Floating Point Instructions: max"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 80, operation = "max", adapter = "direct_packed_u32" }
@@ -323,7 +323,7 @@ ptx_isa_section = "9.7.4.5 Half Precision Floating Point Instructions: neg"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 80, operation = "neg", adapter = "direct_packed_u32" }
@@ -366,7 +366,7 @@ ptx_isa_section = "9.7.4.6 Half Precision Floating Point Instructions: abs"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "bf16x2", native_minimum_sm = 80, operation = "abs", adapter = "direct_packed_u32" }
@@ -408,7 +408,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "fma", adapter = "direct_packed_u32" }
@@ -450,7 +450,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 80, operation = "fma_relu", adapter = "direct_packed_u32" }
@@ -489,7 +489,7 @@ ptx_isa_section = "9.7.4.1 Half Precision Floating Point Instructions: add"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-add"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "add", adapter = "direct_packed_u32" }
@@ -528,7 +528,7 @@ ptx_isa_section = "9.7.4.2 Half Precision Floating Point Instructions: sub"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-sub"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "sub", adapter = "direct_packed_u32" }
@@ -567,7 +567,7 @@ ptx_isa_section = "9.7.4.3 Half Precision Floating Point Instructions: mul"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-mul"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "mul", adapter = "direct_packed_u32" }
@@ -609,7 +609,7 @@ ptx_isa_section = "9.7.4.7 Half Precision Floating Point Instructions: min"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-min"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 80, operation = "min", adapter = "direct_packed_u32" }
@@ -651,7 +651,7 @@ ptx_isa_section = "9.7.4.8 Half Precision Floating Point Instructions: max"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-max"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 80, operation = "max", adapter = "direct_packed_u32" }
@@ -690,7 +690,7 @@ ptx_isa_section = "9.7.4.5 Half Precision Floating Point Instructions: neg"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-neg"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "neg", adapter = "direct_packed_u32" }
@@ -733,7 +733,7 @@ ptx_isa_section = "9.7.4.6 Half Precision Floating Point Instructions: abs"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-abs"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.5", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.5", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.5", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "abs", adapter = "direct_packed_u32" }
@@ -775,7 +775,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "fma_ftz", adapter = "direct_packed_u32" }
@@ -817,7 +817,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "fma_sat", adapter = "direct_packed_u32" }
@@ -859,7 +859,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "4.2", minimum_sm = "sm_75" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 53, operation = "fma_ftz_sat", adapter = "direct_packed_u32" }
@@ -901,7 +901,7 @@ ptx_isa_section = "9.7.4.4 Half Precision Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#half-precision-floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_alu = { format = "f16x2", native_minimum_sm = 80, operation = "fma_ftz_relu", adapter = "direct_packed_u32" }
@@ -940,7 +940,7 @@ ptx_isa_section = "9.7.3.1 Floating Point Instructions: add"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "add", adapter = "direct_packed_u64" }
@@ -979,7 +979,7 @@ ptx_isa_section = "9.7.3.1 Floating Point Instructions: add"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-add"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "add_ftz", adapter = "direct_packed_u64" }
@@ -1018,7 +1018,7 @@ ptx_isa_section = "9.7.3.2 Floating Point Instructions: sub"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "sub", adapter = "direct_packed_u64" }
@@ -1057,7 +1057,7 @@ ptx_isa_section = "9.7.3.2 Floating Point Instructions: sub"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sub"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "sub_ftz", adapter = "direct_packed_u64" }
@@ -1096,7 +1096,7 @@ ptx_isa_section = "9.7.3.3 Floating Point Instructions: mul"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "mul", adapter = "direct_packed_u64" }
@@ -1135,7 +1135,7 @@ ptx_isa_section = "9.7.3.3 Floating Point Instructions: mul"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-mul"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "mul_ftz", adapter = "direct_packed_u64" }
@@ -1174,7 +1174,7 @@ ptx_isa_section = "9.7.3.4 Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "fma", adapter = "direct_packed_u64" }
@@ -1213,7 +1213,7 @@ ptx_isa_section = "9.7.3.4 Floating Point Instructions: fma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-fma"
 lowering = "generated_packed_alu_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "8.6", minimum_sm = "sm_100" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "8.6", minimum_sm = "sm_100" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-12.9-libnvvm-12.9.86-f32x2", minimum_ptx = "8.6", minimum_sm = "sm_100" },
 ]
 packed_alu = { format = "f32x2", native_minimum_sm = 100, operation = "fma_ftz", adapter = "direct_packed_u64" }
diff --git a/cuda-oxide/intrinsics/overlay/packed_atomic.toml b/cuda-oxide/intrinsics/overlay/packed_atomic.toml
index 8b2a9f8c6b..cb3cce7399 100644
--- a/cuda-oxide/intrinsics/overlay/packed_atomic.toml
+++ b/cuda-oxide/intrinsics/overlay/packed_atomic.toml
@@ -34,7 +34,7 @@ ptx_isa_section = "9.7.13.1.1 Atomic reduction and atomic operations: atom"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom"
 lowering = "generated_packed_atomic_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.2", minimum_sm = "sm_70" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.2", minimum_sm = "sm_70" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.2", minimum_sm = "sm_75" },
 ]
 packed_atomic = { format = "f16x2", native_minimum_sm = 60, operation = "add", state_space = "global", ordering = "relaxed", scope = "gpu", rounding = "nearest_even", subnormal = "preserve", atomicity = "per_element", pointer_contract = "mutable_global_u32_aligned4", access_contract = "no_mixed_whole_word_or_non_atomic_access", scope_contract = "racing_atomics_mutually_inclusive", codegen_contract = "exact_native_instruction", return_contract = "old_values_per_element_may_be_noncoherent", adapter = "old_packed_u32", runtime_validation = "unexecuted" }
@@ -72,7 +72,7 @@ ptx_isa_section = "9.7.13.1.1 Atomic reduction and atomic operations: atom"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-atom"
 lowering = "generated_packed_atomic_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.8", minimum_sm = "sm_90" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.8", minimum_sm = "sm_90" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.8", minimum_sm = "sm_90" },
 ]
 packed_atomic = { format = "bf16x2", native_minimum_sm = 90, operation = "add", state_space = "global", ordering = "relaxed", scope = "gpu", rounding = "nearest_even", subnormal = "preserve", atomicity = "per_element", pointer_contract = "mutable_global_u32_aligned4", access_contract = "no_mixed_whole_word_or_non_atomic_access", scope_contract = "racing_atomics_mutually_inclusive", codegen_contract = "exact_native_instruction", return_contract = "old_values_per_element_may_be_noncoherent", adapter = "old_packed_u32", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/packed_conversion.toml b/cuda-oxide/intrinsics/overlay/packed_conversion.toml
index 3c5984fc21..9118f0700d 100644
--- a/cuda-oxide/intrinsics/overlay/packed_conversion.toml
+++ b/cuda-oxide/intrinsics/overlay/packed_conversion.toml
@@ -38,7 +38,7 @@ ptx_isa_section = "9.7.9.22 Data Movement and Conversion Instructions: cvt"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
 lowering = "generated_packed_conversion_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_conversion = { source_format = "f32x2", destination_format = "bf16x2", rounding = "nearest_even", saturation = "none", adapter = "reverse_high_low_operands" }
@@ -80,7 +80,7 @@ ptx_isa_section = "9.7.9.22 Data Movement and Conversion Instructions: cvt"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
 lowering = "generated_packed_conversion_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_conversion = { source_format = "f32x2", destination_format = "f16x2", rounding = "nearest_even", saturation = "none", adapter = "reverse_high_low_operands" }
@@ -122,7 +122,7 @@ ptx_isa_section = "9.7.9.22 Data Movement and Conversion Instructions: cvt"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
 lowering = "generated_packed_conversion_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_conversion = { source_format = "f32x2", destination_format = "f16x2", rounding = "toward_zero", saturation = "none", adapter = "reverse_high_low_operands" }
@@ -164,7 +164,7 @@ ptx_isa_section = "9.7.9.22 Data Movement and Conversion Instructions: cvt"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
 lowering = "generated_packed_conversion_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_conversion = { source_format = "f32x2", destination_format = "f16x2", rounding = "nearest_even", saturation = "relu", adapter = "reverse_high_low_operands" }
@@ -206,7 +206,7 @@ ptx_isa_section = "9.7.9.22 Data Movement and Conversion Instructions: cvt"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
 lowering = "generated_packed_conversion_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_conversion = { source_format = "f32x2", destination_format = "bf16x2", rounding = "nearest_even", saturation = "relu", adapter = "reverse_high_low_operands" }
@@ -248,7 +248,7 @@ ptx_isa_section = "9.7.9.22 Data Movement and Conversion Instructions: cvt"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-cvt"
 lowering = "generated_packed_conversion_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 packed_conversion = { source_format = "f32x2", destination_format = "bf16x2", rounding = "toward_zero", saturation = "none", adapter = "reverse_high_low_operands" }
diff --git a/cuda-oxide/intrinsics/overlay/packed_conversion_fp8.toml b/cuda-oxide/intrinsics/overlay/packed_conversion_fp8.toml
index fdf3d4bece..d643b60468 100644
--- a/cuda-oxide/intrinsics/overlay/packed_conversion_fp8.toml
+++ b/cuda-oxide/intrinsics/overlay/packed_conversion_fp8.toml
@@ -4,7 +4,7 @@ schema = 29
 family = "packed_conversion"
 
 [packed_conversion_fp8]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 destination_formats = ["e4m3x2", "e5m2x2"]
diff --git a/cuda-oxide/intrinsics/overlay/packed_conversion_fp8_f16x2.toml b/cuda-oxide/intrinsics/overlay/packed_conversion_fp8_f16x2.toml
index e4052c017f..abc169c339 100644
--- a/cuda-oxide/intrinsics/overlay/packed_conversion_fp8_f16x2.toml
+++ b/cuda-oxide/intrinsics/overlay/packed_conversion_fp8_f16x2.toml
@@ -8,7 +8,7 @@ schema = 59
 family = "packed_conversion"
 
 [packed_conversion_fp8_f16x2]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 fp8_formats = ["e4m3x2", "e5m2x2"]
diff --git a/cuda-oxide/intrinsics/overlay/prmt.toml b/cuda-oxide/intrinsics/overlay/prmt.toml
index 8e72d74da6..66513a93d2 100644
--- a/cuda-oxide/intrinsics/overlay/prmt.toml
+++ b/cuda-oxide/intrinsics/overlay/prmt.toml
@@ -2,7 +2,7 @@ schema = 28
 family = "prmt"
 
 [prmt]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-prmt"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/register_control.toml b/cuda-oxide/intrinsics/overlay/register_control.toml
index a04b33fcd5..810a7a10d7 100644
--- a/cuda-oxide/intrinsics/overlay/register_control.toml
+++ b/cuda-oxide/intrinsics/overlay/register_control.toml
@@ -36,7 +36,7 @@ ptx_isa_section = "9.7.16.5 Miscellaneous Instructions: setmaxnreg"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "8.0" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "8.0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "8.0" },
 ]
 expected_ptx = { mnemonic = "setmaxnreg", modifiers = ["inc", "sync", "aligned", "u32"], operands = [{ kind = "immediate" }] }
@@ -75,7 +75,7 @@ ptx_isa_section = "9.7.16.5 Miscellaneous Instructions: setmaxnreg"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#miscellaneous-instructions-setmaxnreg"
 lowering = "generated_execution_control"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc-execution-control", minimum_ptx = "8.0" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0-execution-control", minimum_ptx = "8.0" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.73-execution-control", minimum_ptx = "8.0" },
 ]
 expected_ptx = { mnemonic = "setmaxnreg", modifiers = ["dec", "sync", "aligned", "u32"], operands = [{ kind = "immediate" }] }
diff --git a/cuda-oxide/intrinsics/overlay/register_mma.toml b/cuda-oxide/intrinsics/overlay/register_mma.toml
index ad4319bf5a..9a44b84a6c 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.15.5.14 Multiply-and-Accumulate Instruction: mma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
 lowering = "generated_register_mma"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 register_mma = { operation = "multiply", shape = "m16n8k16", accumulator = "f32", a_element = "bf16", b_element = "bf16", a_layout = "row", b_layout = "col", overflow = "not_applicable", participation = "all_warp_lanes_same_instruction_and_qualifiers_no_exited_lanes", adapter = "c4_f32_a4_u32_b2_u32_to_d4_f32", compatibility_source = "existing_stub", runtime_validation = "unexecuted" }
@@ -78,7 +78,7 @@ ptx_isa_section = "9.7.15.5.14 Multiply-and-Accumulate Instruction: mma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
 lowering = "generated_register_mma"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 register_mma = { operation = "multiply", shape = "m16n8k16", accumulator = "f32", a_element = "f16", b_element = "f16", a_layout = "row", b_layout = "col", overflow = "not_applicable", participation = "all_warp_lanes_same_instruction_and_qualifiers_no_exited_lanes", adapter = "c4_f32_a4_u32_b2_u32_to_d4_f32", compatibility_source = "existing_stub", runtime_validation = "unexecuted" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.15.5.14 Multiply-and-Accumulate Instruction: mma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
 lowering = "generated_register_mma"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 register_mma = { operation = "multiply", shape = "m16n8k8", accumulator = "f32", a_element = "tf32", b_element = "tf32", a_layout = "row", b_layout = "col", overflow = "not_applicable", participation = "all_warp_lanes_same_instruction_and_qualifiers_no_exited_lanes", adapter = "c4_f32_a4_u32_b2_u32_to_d4_f32", compatibility_source = "existing_stub", runtime_validation = "unexecuted" }
@@ -160,7 +160,7 @@ ptx_isa_section = "9.7.15.5.14 Multiply-and-Accumulate Instruction: mma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
 lowering = "generated_register_mma"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 register_mma = { operation = "multiply", shape = "m16n8k32", accumulator = "s32", a_element = "s8", b_element = "s8", a_layout = "row", b_layout = "col", overflow = "wrapping", participation = "all_warp_lanes_same_instruction_and_qualifiers_no_exited_lanes", adapter = "c4_i32_a4_u32_b2_u32_to_d4_i32", compatibility_source = "existing_stub", runtime_validation = "unexecuted" }
@@ -201,7 +201,7 @@ ptx_isa_section = "9.7.15.5.14 Multiply-and-Accumulate Instruction: mma"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#warp-level-matrix-instructions-mma"
 lowering = "generated_register_mma"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "7.0", minimum_sm = "sm_80" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "7.0", minimum_sm = "sm_80" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "7.0", minimum_sm = "sm_80" },
 ]
 register_mma = { operation = "multiply", shape = "m8n8k4", accumulator = "f64", a_element = "f64", b_element = "f64", a_layout = "row", b_layout = "col", overflow = "not_applicable", participation = "all_warp_lanes_same_instruction_and_qualifiers_no_exited_lanes", adapter = "c2_f64_a1_f64_b1_f64_to_d2_f64", compatibility_source = "existing_stub", runtime_validation = "unexecuted" }
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_ampere_float.toml b/cuda-oxide/intrinsics/overlay/register_mma_ampere_float.toml
index cc1f2bb4b5..a348e28319 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_ampere_float.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_ampere_float.toml
@@ -2,7 +2,7 @@ schema = 49
 family = "register_mma"
 
 [register_mma_ampere_float]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 first_abi_id = "i0520"
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_b1.toml b/cuda-oxide/intrinsics/overlay/register_mma_b1.toml
index a3150e9c3a..f13cfa7d37 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_b1.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_b1.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "register_mma"
 
 [register_mma_b1]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f16.toml b/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f16.toml
index 3637d3a6bf..8cbe57e2c2 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f16.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f16.toml
@@ -4,7 +4,7 @@ schema = 47
 family = "register_mma"
 
 [register_mma_f8f6f4_f16]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 first_abi_id = "i0479"
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f32.toml b/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f32.toml
index 88947affe0..10ca2b3aea 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f32.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_f8f6f4_f32.toml
@@ -4,7 +4,7 @@ schema = 46
 family = "register_mma"
 
 [register_mma_f8f6f4_f32]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 first_abi_id = "i0454"
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_fp8.toml b/cuda-oxide/intrinsics/overlay/register_mma_fp8.toml
index daf6660b65..f72494de5e 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_fp8.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_fp8.toml
@@ -2,7 +2,7 @@ schema = 48
 family = "register_mma"
 
 [register_mma_fp8]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 first_abi_id = "i0504"
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_int4_m16.toml b/cuda-oxide/intrinsics/overlay/register_mma_int4_m16.toml
index 0605e28b7d..55a62f59ad 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_int4_m16.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_int4_m16.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "register_mma"
 
 [register_mma_int4]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_int4_m8.toml b/cuda-oxide/intrinsics/overlay/register_mma_int4_m8.toml
index f731c77106..fcbc2c5c38 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_int4_m8.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_int4_m8.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "register_mma"
 
 [register_mma_int4]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_int8.toml b/cuda-oxide/intrinsics/overlay/register_mma_int8.toml
index 0d838a065c..df0c32be30 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_int8.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_int8.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "register_mma"
 
 [register_mma_int8]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_int8_m8.toml b/cuda-oxide/intrinsics/overlay/register_mma_int8_m8.toml
index dd907c0986..2cd505010d 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_int8_m8.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_int8_m8.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "register_mma"
 
 [register_mma_int8]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/register_mma_mxf8f6f4_f32.toml b/cuda-oxide/intrinsics/overlay/register_mma_mxf8f6f4_f32.toml
index 9c646926db..6b5f6ce31f 100644
--- a/cuda-oxide/intrinsics/overlay/register_mma_mxf8f6f4_f32.toml
+++ b/cuda-oxide/intrinsics/overlay/register_mma_mxf8f6f4_f32.toml
@@ -4,7 +4,7 @@ schema = 60
 family = "register_mma"
 
 [register_mma_mxf8f6f4_f32]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.73"
 runtime_validation = "unexecuted"
 first_abi_id = "i0858"
diff --git a/cuda-oxide/intrinsics/overlay/scalar_arithmetic.toml b/cuda-oxide/intrinsics/overlay/scalar_arithmetic.toml
index 84b81f9adb..61b8c983bb 100644
--- a/cuda-oxide/intrinsics/overlay/scalar_arithmetic.toml
+++ b/cuda-oxide/intrinsics/overlay/scalar_arithmetic.toml
@@ -4,7 +4,7 @@ schema = 45
 family = "scalar_arithmetic"
 
 [scalar_arithmetic]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-scalar-arithmetic"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/scalar_conversion.toml b/cuda-oxide/intrinsics/overlay/scalar_conversion.toml
index 2c6cdc3a26..0d515ff7dd 100644
--- a/cuda-oxide/intrinsics/overlay/scalar_conversion.toml
+++ b/cuda-oxide/intrinsics/overlay/scalar_conversion.toml
@@ -2,7 +2,7 @@ schema = 43
 family = "scalar_conversion"
 
 [scalar_conversion]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-scalar-conversion"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/scalar_math.toml b/cuda-oxide/intrinsics/overlay/scalar_math.toml
index 88c5b748b4..d9a26b4cb0 100644
--- a/cuda-oxide/intrinsics/overlay/scalar_math.toml
+++ b/cuda-oxide/intrinsics/overlay/scalar_math.toml
@@ -4,7 +4,7 @@ schema = 58
 family = "scalar_math"
 
 [scalar_math]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-scalar-math"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_fp8_f32.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_fp8_f32.toml
index d9ab0a0653..cfd31a1005 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_fp8_f32.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_fp8_f32.toml
@@ -12,7 +12,7 @@ family = "sparse_mma"
 # rather than the V13.3.33 build the other shards were recorded against; the
 # evidence file name carries the same release identity as its backend_version.
 [sparse_mma_fp8_f32]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.73"
 runtime_validation = "unexecuted"
 a_elements = ["e4m3", "e5m2"]
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k128.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k128.toml
index 6042385774..f4839429c0 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k128.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k128.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "standard"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k64.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k64.toml
index 1c20b67367..298757b0dd 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k64.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_int4_m16n8k64.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "standard"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_int8.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_int8.toml
index 14627b45a5..5c25899ae5 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_int8.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_int8.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "standard"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_int8_m16n8k64.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_int8_m16n8k64.toml
index b9eace51c3..a37b8dbc32 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_int8_m16n8k64.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_int8_m16n8k64.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "standard"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_ampere_float.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_ampere_float.toml
index 81c3e6a775..157f8d1819 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_ampere_float.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_ampere_float.toml
@@ -4,7 +4,7 @@ schema = 63
 family = "sparse_mma"
 
 [sparse_mma_ordered_ampere_float]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "executed"
 
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f16.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f16.toml
index 7237dc8afb..8bbecd8cea 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f16.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f16.toml
@@ -4,7 +4,7 @@ schema = 50
 family = "sparse_mma"
 
 [sparse_mma_f8f6f4_f16]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 first_abi_id = "i0525"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f32.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f32.toml
index a43e8116a3..7e4be60aec 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f32.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_f8f6f4_f32.toml
@@ -4,7 +4,7 @@ schema = 27
 family = "sparse_mma"
 
 [sparse_mma_f8f6f4_f32]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 a_elements = ["e2m1", "e2m3", "e3m2", "e4m3", "e5m2"]
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k128.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k128.toml
index d089746eaf..a22459c80e 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k128.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k128.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "ordered"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k64.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k64.toml
index 5717ba3cbc..bad478604a 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k64.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int4_m16n8k64.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "ordered"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8.toml
index 0632380c5d..c73024c6da 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "ordered"
diff --git a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8_m16n8k64.toml b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8_m16n8k64.toml
index ef1bc2a735..b602fc6bef 100644
--- a/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8_m16n8k64.toml
+++ b/cuda-oxide/intrinsics/overlay/sparse_mma_ordered_int8_m16n8k64.toml
@@ -4,7 +4,7 @@ schema = 26
 family = "sparse_mma"
 
 [sparse_mma_integer]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 metadata = "ordered"
diff --git a/cuda-oxide/intrinsics/overlay/sreg_special.toml b/cuda-oxide/intrinsics/overlay/sreg_special.toml
index 0b3ec92c43..cbee601fd2 100644
--- a/cuda-oxide/intrinsics/overlay/sreg_special.toml
+++ b/cuda-oxide/intrinsics/overlay/sreg_special.toml
@@ -4,7 +4,7 @@ schema = 32
 family = "sreg"
 
 [special_registers]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33"
 runtime_validation = "unexecuted"
 registers = [
diff --git a/cuda-oxide/intrinsics/overlay/stmatrix.toml b/cuda-oxide/intrinsics/overlay/stmatrix.toml
index c71b3d3821..3c82b2aeed 100644
--- a/cuda-oxide/intrinsics/overlay/stmatrix.toml
+++ b/cuda-oxide/intrinsics/overlay/stmatrix.toml
@@ -2,7 +2,7 @@ schema = 35
 family = "stmatrix"
 
 [stmatrix]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-stmatrix"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/sync.toml b/cuda-oxide/intrinsics/overlay/sync.toml
index 9d00217b75..7024711e22 100644
--- a/cuda-oxide/intrinsics/overlay/sync.toml
+++ b/cuda-oxide/intrinsics/overlay/sync.toml
@@ -39,7 +39,7 @@ ptx_isa_section = "9.7.14.1 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-barrier"
 lowering = "generated_sync_threads"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "3.2", minimum_sm = "sm_20" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "3.2", minimum_sm = "sm_20" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_sm = "sm_75" },
 ]
 expected_ptx = { mnemonic = "bar", modifiers = ["sync"], operands = [{ kind = "exact", value = "0" }] }
diff --git a/cuda-oxide/intrinsics/overlay/tcgen05.toml b/cuda-oxide/intrinsics/overlay/tcgen05.toml
index d8a616b66a..ba328d17e3 100644
--- a/cuda-oxide/intrinsics/overlay/tcgen05.toml
+++ b/cuda-oxide/intrinsics/overlay/tcgen05.toml
@@ -2,19 +2,19 @@ schema = 57
 family = "tcgen05"
 
 [tcgen05]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05"
-cp_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+cp_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 cp_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05-cp"
-ld_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+ld_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 ld_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05-ld"
-st_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+st_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 st_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05-st"
-offset_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+offset_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 offset_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05-ldst-offset"
-control_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+control_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 control_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05-control"
-mma_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-tcgen05-mma"
+mma_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-tcgen05-mma"
 mma_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-tcgen05-mma"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/threadfence.toml b/cuda-oxide/intrinsics/overlay/threadfence.toml
index 6cdf1dafcd..e4198c41f7 100644
--- a/cuda-oxide/intrinsics/overlay/threadfence.toml
+++ b/cuda-oxide/intrinsics/overlay/threadfence.toml
@@ -2,7 +2,7 @@ schema = 34
 family = "sync"
 
 [threadfence]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-threadfence"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/tma.toml b/cuda-oxide/intrinsics/overlay/tma.toml
index b9b182c98b..fc68674477 100644
--- a/cuda-oxide/intrinsics/overlay/tma.toml
+++ b/cuda-oxide/intrinsics/overlay/tma.toml
@@ -4,9 +4,9 @@ schema = 62
 family = "tma"
 
 [tma]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-tma"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-tma"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.73-tma"
-reduce_llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-tma-reduce"
+reduce_llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-tma-reduce"
 reduce_libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.73-tma-reduce"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/overlay/warp_barrier.toml b/cuda-oxide/intrinsics/overlay/warp_barrier.toml
index 54f39e21c5..483968a23e 100644
--- a/cuda-oxide/intrinsics/overlay/warp_barrier.toml
+++ b/cuda-oxide/intrinsics/overlay/warp_barrier.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.14.2 Parallel Synchronization and Communication Instructi
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#parallel-synchronization-and-communication-instructions-bar-warp-sync"
 lowering = "generated_warp_barrier"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_barrier = { participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", adapter = "direct_member_mask", mask_encoding = "register_or_immediate", memory_ordering = "participating_lanes" }
diff --git a/cuda-oxide/intrinsics/overlay/warp_shuffle.toml b/cuda-oxide/intrinsics/overlay/warp_shuffle.toml
index 00f6c7578c..a985e531ae 100644
--- a/cuda-oxide/intrinsics/overlay/warp_shuffle.toml
+++ b/cuda-oxide/intrinsics/overlay/warp_shuffle.toml
@@ -37,7 +37,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "idx", value_kind = "i32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 31, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -78,7 +78,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "bfly", value_kind = "i32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 31, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -119,7 +119,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "down", value_kind = "i32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 31, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -160,7 +160,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "up", value_kind = "i32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 0, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -201,7 +201,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "idx", value_kind = "f32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 31, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -242,7 +242,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "bfly", value_kind = "f32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 31, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -283,7 +283,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "down", value_kind = "f32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 31, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -324,7 +324,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "typed_nvvm", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "typed_nvvm", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "up", value_kind = "f32", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_insert_clamp", clamp = 0, lane_encoding = "register_or_immediate", mask_encoding = "register_or_immediate" }
@@ -362,7 +362,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle_i64_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "idx", value_kind = "i64", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_split_i64_low_high_b32_insert_clamp_reassemble", clamp = 31, lane_encoding = "register_only", mask_encoding = "register_only" }
@@ -400,7 +400,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle_i64_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "bfly", value_kind = "i64", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_split_i64_low_high_b32_insert_clamp_reassemble", clamp = 31, lane_encoding = "register_only", mask_encoding = "register_only" }
@@ -438,7 +438,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle_i64_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "down", value_kind = "i64", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_split_i64_low_high_b32_insert_clamp_reassemble", clamp = 31, lane_encoding = "register_only", mask_encoding = "register_only" }
@@ -476,7 +476,7 @@ ptx_isa_section = "9.7.9.6 Data Movement and Conversion Instructions: shfl.sync"
 ptx_isa_url = "https://docs.nvidia.com/cuda/parallel-thread-execution/#data-movement-and-conversion-instructions-shfl-sync"
 lowering = "generated_warp_shuffle_i64_inline_ptx"
 backend_lowerings = [
-    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.0-16696adc", minimum_ptx = "6.0", minimum_sm = "sm_30" },
+    { backend = "llvm_nvptx", mechanism = "inline_ptx", evidence_profile = "rust-llvm-23.1.3-aab11ee0", minimum_ptx = "6.0", minimum_sm = "sm_30" },
     { backend = "lib_nvvm", mechanism = "inline_ptx", evidence_profile = "cuda-13.3-libnvvm-13.3.33", minimum_ptx = "6.0", minimum_sm = "sm_75" },
 ]
 warp_shuffle = { mode = "up", value_kind = "i64", participation = "executing_lane_named_all_named_lanes_same_instruction_and_mask", legacy_pre_sm70 = "all_named_lanes_converged_and_only_named_lanes_active", source_lane = "in_range_source_active_and_named_out_of_range_copies_self", adapter = "mask_value_lane_or_delta_split_i64_low_high_b32_insert_clamp_reassemble", clamp = 0, lane_encoding = "register_only", mask_encoding = "register_only" }
diff --git a/cuda-oxide/intrinsics/overlay/wgmma_control.toml b/cuda-oxide/intrinsics/overlay/wgmma_control.toml
index 246a9dae84..e9e44c3014 100644
--- a/cuda-oxide/intrinsics/overlay/wgmma_control.toml
+++ b/cuda-oxide/intrinsics/overlay/wgmma_control.toml
@@ -4,7 +4,7 @@ schema = 40
 family = "wgmma_control"
 
 [wgmma_controls]
-llvm_evidence_profile = "rust-llvm-23.1.0-16696adc-wgmma-control"
+llvm_evidence_profile = "rust-llvm-23.1.3-aab11ee0-wgmma-control"
 libnvvm_evidence_profile = "cuda-13.3-libnvvm-13.3.33-wgmma-control"
 runtime_validation = "unexecuted"
 
diff --git a/cuda-oxide/intrinsics/probes/abs_bf16x2.ll b/cuda-oxide/intrinsics/probes/abs_bf16x2.ll
index c759d6494a..e4f90dbf4f 100644
--- a/cuda-oxide/intrinsics/probes/abs_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/abs_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/abs_f16x2.ll b/cuda-oxide/intrinsics/probes/abs_f16x2.ll
index f79af7cbb5..d282bddf00 100644
--- a/cuda-oxide/intrinsics/probes/abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/active_mask.ll b/cuda-oxide/intrinsics/probes/active_mask.ll
index 3c522a8933..cfb28da0a0 100644
--- a/cuda-oxide/intrinsics/probes/active_mask.ll
+++ b/cuda-oxide/intrinsics/probes/active_mask.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_bf16x2.ll b/cuda-oxide/intrinsics/probes/add_bf16x2.ll
index c3ded6e68a..9e8b4b623c 100644
--- a/cuda-oxide/intrinsics/probes/add_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/add_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_f16x2.ll b/cuda-oxide/intrinsics/probes/add_f16x2.ll
index 899682ab2f..9c9be96eed 100644
--- a/cuda-oxide/intrinsics/probes/add_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/add_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_f32x2.ll b/cuda-oxide/intrinsics/probes/add_f32x2.ll
index 5642fd65d9..8bd18c88a6 100644
--- a/cuda-oxide/intrinsics/probes/add_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/add_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_ftz_f32x2.ll b/cuda-oxide/intrinsics/probes/add_ftz_f32x2.ll
index 2cf4f23b8c..15b70e709e 100644
--- a/cuda-oxide/intrinsics/probes/add_ftz_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/add_ftz_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rm_f32.ll b/cuda-oxide/intrinsics/probes/add_rm_f32.ll
index 22090a5d47..6db9ffcbc0 100644
--- a/cuda-oxide/intrinsics/probes/add_rm_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rm_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rm_f64.ll b/cuda-oxide/intrinsics/probes/add_rm_f64.ll
index c142dc8d38..969c43674e 100644
--- a/cuda-oxide/intrinsics/probes/add_rm_f64.ll
+++ b/cuda-oxide/intrinsics/probes/add_rm_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rm_ftz_f32.ll b/cuda-oxide/intrinsics/probes/add_rm_ftz_f32.ll
index 07a1b1b8a9..aa286f0f62 100644
--- a/cuda-oxide/intrinsics/probes/add_rm_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rm_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rm_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rm_ftz_sat_f32.ll
index f42e2f9a98..b6b6565745 100644
--- a/cuda-oxide/intrinsics/probes/add_rm_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rm_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rm_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rm_sat_f32.ll
index 25e4199acd..a03d96f4a1 100644
--- a/cuda-oxide/intrinsics/probes/add_rm_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rm_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rn_f32.ll b/cuda-oxide/intrinsics/probes/add_rn_f32.ll
index a462c088b9..1c618b8bff 100644
--- a/cuda-oxide/intrinsics/probes/add_rn_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rn_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rn_f64.ll b/cuda-oxide/intrinsics/probes/add_rn_f64.ll
index 3c07852ba2..cdc0104e4b 100644
--- a/cuda-oxide/intrinsics/probes/add_rn_f64.ll
+++ b/cuda-oxide/intrinsics/probes/add_rn_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rn_ftz_f32.ll b/cuda-oxide/intrinsics/probes/add_rn_ftz_f32.ll
index 65b6b22de1..5af9974339 100644
--- a/cuda-oxide/intrinsics/probes/add_rn_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rn_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rn_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rn_ftz_sat_f32.ll
index bb5f6b1bfa..12355cdf98 100644
--- a/cuda-oxide/intrinsics/probes/add_rn_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rn_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rn_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rn_sat_f32.ll
index 0b0a8722d0..31e4fc10a6 100644
--- a/cuda-oxide/intrinsics/probes/add_rn_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rn_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rp_f32.ll b/cuda-oxide/intrinsics/probes/add_rp_f32.ll
index e9831836d4..2e8a575e27 100644
--- a/cuda-oxide/intrinsics/probes/add_rp_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rp_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rp_f64.ll b/cuda-oxide/intrinsics/probes/add_rp_f64.ll
index a0e14b6d2d..5e098e67bc 100644
--- a/cuda-oxide/intrinsics/probes/add_rp_f64.ll
+++ b/cuda-oxide/intrinsics/probes/add_rp_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rp_ftz_f32.ll b/cuda-oxide/intrinsics/probes/add_rp_ftz_f32.ll
index 4ef74410fc..6daf152ec6 100644
--- a/cuda-oxide/intrinsics/probes/add_rp_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rp_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rp_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rp_ftz_sat_f32.ll
index 37afea0e75..731fb3e96a 100644
--- a/cuda-oxide/intrinsics/probes/add_rp_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rp_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rp_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rp_sat_f32.ll
index ac90be34ac..4922207ad2 100644
--- a/cuda-oxide/intrinsics/probes/add_rp_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rp_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rz_f32.ll b/cuda-oxide/intrinsics/probes/add_rz_f32.ll
index dcd877842b..1d6bc06f27 100644
--- a/cuda-oxide/intrinsics/probes/add_rz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rz_f64.ll b/cuda-oxide/intrinsics/probes/add_rz_f64.ll
index 26ca838cfe..66bfa0676a 100644
--- a/cuda-oxide/intrinsics/probes/add_rz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/add_rz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rz_ftz_f32.ll b/cuda-oxide/intrinsics/probes/add_rz_ftz_f32.ll
index e4b909f948..de5dedd64b 100644
--- a/cuda-oxide/intrinsics/probes/add_rz_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rz_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rz_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rz_ftz_sat_f32.ll
index 01e08bb065..9182470492 100644
--- a/cuda-oxide/intrinsics/probes/add_rz_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rz_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/add_rz_sat_f32.ll b/cuda-oxide/intrinsics/probes/add_rz_sat_f32.ll
index 350187eb9a..cceab23750 100644
--- a/cuda-oxide/intrinsics/probes/add_rz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/add_rz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/all_sync.ll b/cuda-oxide/intrinsics/probes/all_sync.ll
index 54df3f0fc5..cb66ad10f4 100644
--- a/cuda-oxide/intrinsics/probes/all_sync.ll
+++ b/cuda-oxide/intrinsics/probes/all_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/any_sync.ll b/cuda-oxide/intrinsics/probes/any_sync.ll
index c62e870fbb..97b18cd57a 100644
--- a/cuda-oxide/intrinsics/probes/any_sync.ll
+++ b/cuda-oxide/intrinsics/probes/any_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ballot_sync.ll b/cuda-oxide/intrinsics/probes/ballot_sync.ll
index 1c0a90549d..96081557da 100644
--- a/cuda-oxide/intrinsics/probes/ballot_sync.ll
+++ b/cuda-oxide/intrinsics/probes/ballot_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive.ll b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive.ll
index 158357505b..304516396b 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_aligned.ll b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_aligned.ll
index 901fffaf32..cfc169549d 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_aligned.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_aligned.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed.ll b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed.ll
index e85e359968..c03df80219 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed_aligned.ll b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed_aligned.ll
index 9d983a9cd9..08e8e7e515 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed_aligned.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cluster_arrive_relaxed_aligned.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cluster_wait.ll b/cuda-oxide/intrinsics/probes/barrier_cluster_wait.ll
index b7bfb2f9c2..4c91be89ba 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cluster_wait.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cluster_wait.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cluster_wait_aligned.ll b/cuda-oxide/intrinsics/probes/barrier_cluster_wait_aligned.ll
index 1ce1ef8a24..ad7e518834 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cluster_wait_aligned.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cluster_wait_aligned.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cta_arrive.ll b/cuda-oxide/intrinsics/probes/barrier_cta_arrive.ll
index 8c37c624cb..ccaed8aec9 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cta_arrive.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cta_arrive.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cta_arrive_aligned.ll b/cuda-oxide/intrinsics/probes/barrier_cta_arrive_aligned.ll
index 5e9555a1b5..97fb497976 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cta_arrive_aligned.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cta_arrive_aligned.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cta_sync.ll b/cuda-oxide/intrinsics/probes/barrier_cta_sync.ll
index 026e7984d5..24e0cf60ac 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cta_sync.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cta_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/barrier_cta_sync_aligned.ll b/cuda-oxide/intrinsics/probes/barrier_cta_sync_aligned.ll
index 6309af77f5..b1d298dd80 100644
--- a/cuda-oxide/intrinsics/probes/barrier_cta_sync_aligned.ll
+++ b/cuda-oxide/intrinsics/probes/barrier_cta_sync_aligned.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/block_dim_x.ll b/cuda-oxide/intrinsics/probes/block_dim_x.ll
index eec87d7cb2..ad845532eb 100644
--- a/cuda-oxide/intrinsics/probes/block_dim_x.ll
+++ b/cuda-oxide/intrinsics/probes/block_dim_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/block_dim_y.ll b/cuda-oxide/intrinsics/probes/block_dim_y.ll
index 78c06ddb31..3d78afddb1 100644
--- a/cuda-oxide/intrinsics/probes/block_dim_y.ll
+++ b/cuda-oxide/intrinsics/probes/block_dim_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/block_dim_z.ll b/cuda-oxide/intrinsics/probes/block_dim_z.ll
index fc09e16013..7048c46326 100644
--- a/cuda-oxide/intrinsics/probes/block_dim_z.ll
+++ b/cuda-oxide/intrinsics/probes/block_dim_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/block_idx_x.ll b/cuda-oxide/intrinsics/probes/block_idx_x.ll
index 443bbc24f2..2066c67b2a 100644
--- a/cuda-oxide/intrinsics/probes/block_idx_x.ll
+++ b/cuda-oxide/intrinsics/probes/block_idx_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/block_idx_y.ll b/cuda-oxide/intrinsics/probes/block_idx_y.ll
index a7099664d3..56b24a2ca6 100644
--- a/cuda-oxide/intrinsics/probes/block_idx_y.ll
+++ b/cuda-oxide/intrinsics/probes/block_idx_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/block_idx_z.ll b/cuda-oxide/intrinsics/probes/block_idx_z.ll
index 9953c01142..69ec9340ab 100644
--- a/cuda-oxide/intrinsics/probes/block_idx_z.ll
+++ b/cuda-oxide/intrinsics/probes/block_idx_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/breakpoint.ll b/cuda-oxide/intrinsics/probes/breakpoint.ll
index 032084ab2f..fee6f3fa0c 100644
--- a/cuda-oxide/intrinsics/probes/breakpoint.ll
+++ b/cuda-oxide/intrinsics/probes/breakpoint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_x.ll b/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_x.ll
index 3dae251d11..4920f565fa 100644
--- a/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_x.ll
+++ b/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_y.ll b/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_y.ll
index 77a2459449..d3dc26d79b 100644
--- a/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_y.ll
+++ b/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_z.ll b/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_z.ll
index 36566c4490..418a1c73fa 100644
--- a/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_z.ll
+++ b/cuda-oxide/intrinsics/probes/clc_query_get_first_ctaid_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clc_query_is_canceled.ll b/cuda-oxide/intrinsics/probes/clc_query_is_canceled.ll
index 9f0ada4b39..aec4b27149 100644
--- a/cuda-oxide/intrinsics/probes/clc_query_is_canceled.ll
+++ b/cuda-oxide/intrinsics/probes/clc_query_is_canceled.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clc_try_cancel.ll b/cuda-oxide/intrinsics/probes/clc_try_cancel.ll
index f89bfecf25..a44a1cfa20 100644
--- a/cuda-oxide/intrinsics/probes/clc_try_cancel.ll
+++ b/cuda-oxide/intrinsics/probes/clc_try_cancel.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clc_try_cancel_multicast.ll b/cuda-oxide/intrinsics/probes/clc_try_cancel_multicast.ll
index fa96d77c86..38ffb0ce73 100644
--- a/cuda-oxide/intrinsics/probes/clc_try_cancel_multicast.ll
+++ b/cuda-oxide/intrinsics/probes/clc_try_cancel_multicast.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clock.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/clock.lib_nvvm.ll
index 8ac5ec9072..56dfe3d6b4 100644
--- a/cuda-oxide/intrinsics/probes/clock.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/clock.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clock.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/clock.llvm_nvptx.ll
index 1673854a38..aed88470f6 100644
--- a/cuda-oxide/intrinsics/probes/clock.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/clock.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clock64.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/clock64.lib_nvvm.ll
index 726cf4829a..f51209ab7f 100644
--- a/cuda-oxide/intrinsics/probes/clock64.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/clock64.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/clock64.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/clock64.llvm_nvptx.ll
index 77d94474c2..3148c69953 100644
--- a/cuda-oxide/intrinsics/probes/clock64.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/clock64.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_block_count.ll b/cuda-oxide/intrinsics/probes/cluster_block_count.ll
index c77af0cb8b..39fcd3f775 100644
--- a/cuda-oxide/intrinsics/probes/cluster_block_count.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_block_count.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_block_idx_x.ll b/cuda-oxide/intrinsics/probes/cluster_block_idx_x.ll
index f3a4ddd821..9384df8463 100644
--- a/cuda-oxide/intrinsics/probes/cluster_block_idx_x.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_block_idx_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_block_idx_y.ll b/cuda-oxide/intrinsics/probes/cluster_block_idx_y.ll
index 7f21d0a12e..72edce8ff2 100644
--- a/cuda-oxide/intrinsics/probes/cluster_block_idx_y.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_block_idx_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_block_idx_z.ll b/cuda-oxide/intrinsics/probes/cluster_block_idx_z.ll
index 91ad3fd6cd..d3ee68d839 100644
--- a/cuda-oxide/intrinsics/probes/cluster_block_idx_z.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_block_idx_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_block_rank.ll b/cuda-oxide/intrinsics/probes/cluster_block_rank.ll
index cfb6a0ea65..b5e1ebc766 100644
--- a/cuda-oxide/intrinsics/probes/cluster_block_rank.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_block_rank.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_dim_x.ll b/cuda-oxide/intrinsics/probes/cluster_dim_x.ll
index 38705818f8..dcb424b2de 100644
--- a/cuda-oxide/intrinsics/probes/cluster_dim_x.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_dim_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_dim_y.ll b/cuda-oxide/intrinsics/probes/cluster_dim_y.ll
index 45853409df..08cb319243 100644
--- a/cuda-oxide/intrinsics/probes/cluster_dim_y.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_dim_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_dim_z.ll b/cuda-oxide/intrinsics/probes/cluster_dim_z.ll
index 9b9e9ab058..0f89613939 100644
--- a/cuda-oxide/intrinsics/probes/cluster_dim_z.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_dim_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_grid_dim_x.ll b/cuda-oxide/intrinsics/probes/cluster_grid_dim_x.ll
index 8b304b0807..d3007ef56d 100644
--- a/cuda-oxide/intrinsics/probes/cluster_grid_dim_x.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_grid_dim_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_grid_dim_y.ll b/cuda-oxide/intrinsics/probes/cluster_grid_dim_y.ll
index 5146b69270..be7a7a08aa 100644
--- a/cuda-oxide/intrinsics/probes/cluster_grid_dim_y.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_grid_dim_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_grid_dim_z.ll b/cuda-oxide/intrinsics/probes/cluster_grid_dim_z.ll
index ad3e521abf..51b8879cae 100644
--- a/cuda-oxide/intrinsics/probes/cluster_grid_dim_z.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_grid_dim_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_idx_x.ll b/cuda-oxide/intrinsics/probes/cluster_idx_x.ll
index 28a37d350f..b0f443649b 100644
--- a/cuda-oxide/intrinsics/probes/cluster_idx_x.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_idx_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_idx_y.ll b/cuda-oxide/intrinsics/probes/cluster_idx_y.ll
index b7265dc311..69fc747df9 100644
--- a/cuda-oxide/intrinsics/probes/cluster_idx_y.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_idx_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cluster_idx_z.ll b/cuda-oxide/intrinsics/probes/cluster_idx_z.ll
index 2432fc25b3..c72d22ab0d 100644
--- a/cuda-oxide/intrinsics/probes/cluster_idx_z.ll
+++ b/cuda-oxide/intrinsics/probes/cluster_idx_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cos_approx_f32.ll b/cuda-oxide/intrinsics/probes/cos_approx_f32.ll
index 463c30ec0e..0e3115de83 100644
--- a/cuda-oxide/intrinsics/probes/cos_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cos_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cos_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/cos_approx_ftz_f32.ll
index 0cc4f4c0fe..6e16ebc12f 100644
--- a/cuda-oxide/intrinsics/probes/cos_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cos_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_commit_group.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_commit_group.ll
index 57e937b295..405c6b104e 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_commit_group.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_commit_group.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2.ll
index 0e945a5908..a1bfe4e31c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2_cache_hint.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2_cache_hint.ll
index 80d5f1d28a..72da0da001 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2_cache_hint.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_1d_l2_cache_hint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2.ll
index 2e92cb76af..f9247ae03f 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2_cache_hint.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2_cache_hint.ll
index c2c54d7f6b..a49fab07b3 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2_cache_hint.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_2d_l2_cache_hint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2.ll
index 05aa3cf7ea..734a7ae722 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2_cache_hint.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2_cache_hint.ll
index 78afd6cecf..a927796f63 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2_cache_hint.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_3d_l2_cache_hint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2.ll
index bf5b50969a..51a44d497d 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2_cache_hint.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2_cache_hint.ll
index a6a48ff86d..5962e7a5a3 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2_cache_hint.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_4d_l2_cache_hint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2.ll
index b6d31f8453..b4e0d8cfd5 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2_cache_hint.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2_cache_hint.ll
index e9f51ee313..42ba75f968 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2_cache_hint.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_5d_l2_cache_hint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2.ll
index 979a66ff62..0c97d6057b 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint.ll
index 388124aedd..dcf767ab90 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_prefetch_tensor_gather4_2d_l2_cache_hint.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_g2s.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_g2s.ll
index cf4946e20f..03600b606f 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_g2s.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_g2s.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_s2g.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_s2g.ll
index 1f759f4e24..899b146871 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_s2g.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_1d_s2g.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s.ll
index 805fb69ca3..1f7da55135 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast.ll
index 37ed46fd0d..c990b39f80 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast_cg2.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast_cg2.ll
index 3c74138fa2..b2266ee055 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_g2s_multicast_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_s2g.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_s2g.ll
index 3a3aaf72c5..8b9b870c67 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_s2g.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_2d_s2g.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_g2s.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_g2s.ll
index 2bddcc5431..39f9357b2c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_g2s.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_g2s.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_s2g.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_s2g.ll
index d49e54e46d..83dd948659 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_s2g.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_3d_s2g.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_g2s.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_g2s.ll
index 171dcd3f52..f726d34eb9 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_g2s.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_g2s.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_s2g.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_s2g.ll
index 4a92de939e..bf08d0169b 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_s2g.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_4d_s2g.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_g2s.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_g2s.ll
index fdd9db39f5..7515b63d4c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_g2s.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_g2s.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_s2g.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_s2g.ll
index 57ba0b709b..8b3c1839f9 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_s2g.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_5d_s2g.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_3d.ll
index 73c0167e00..5c3dedff9e 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_4d.ll
index c535f257e9..96feaf9b06 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_5d.ll
index 10a45a3fd0..045c4388c4 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_1d.ll
index 3264bac5b8..35062c7b02 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_2d.ll
index c31a4e0abe..a529f2b873 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_3d.ll
index c670cf96c5..d2977dfc28 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_4d.ll
index b9c9b2f995..e18e0ecee3 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_5d.ll
index f97b3728ad..9471b296a5 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_add_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_3d.ll
index 361cd7a4a9..b7ea799c65 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_4d.ll
index cfcb0f8958..cd741e34cd 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_5d.ll
index dce921c5da..91d6f48365 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_1d.ll
index 8411827e33..07e0943450 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_2d.ll
index 66a8bc4819..ca6edd125f 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_3d.ll
index ac83efb656..0aa4d4b10c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_4d.ll
index 2abbcf4bb5..0b1a2fe822 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_5d.ll
index b9e69dd89c..3806624727 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_and_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_3d.ll
index 7d459c971e..a922f25e39 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_4d.ll
index d721c627c3..0252d6f7e0 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_5d.ll
index 289fd56aac..1b81e77aeb 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_1d.ll
index f716e7af93..d5c2922c22 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_2d.ll
index 9434682d63..1486b3d709 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_3d.ll
index d9248a8d04..41dc2529ea 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_4d.ll
index a287e08da9..ae03b1fceb 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_5d.ll
index a901151478..9619620a5c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_dec_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_3d.ll
index 78f91df966..2dd47ef1bc 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_4d.ll
index 9ab6cbe102..b2793b839a 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_5d.ll
index 9a94956802..f59d0ab3dd 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_1d.ll
index e88074227e..bb49fcd1a4 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_2d.ll
index 40c298537e..a6c45925a5 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_3d.ll
index 1817063a88..c885e75016 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_4d.ll
index d8a420acc2..6a1a53cdaf 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_5d.ll
index 3c6f129025..9a84e9a095 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_inc_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_3d.ll
index 0873d0284c..d1ed727d67 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_4d.ll
index a56a952adf..db43ebc6d7 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_5d.ll
index b53d95e547..c65c97f089 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_1d.ll
index 0f8ece64b8..180c30a72c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_2d.ll
index eb2657563d..9e49134245 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_3d.ll
index 8fb7afa4f3..6888b3f9fc 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_4d.ll
index 73391f8c73..2727eb9958 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_5d.ll
index f6786f5755..b26850ff5f 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_max_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_3d.ll
index 136abe2ccd..4a303f2f91 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_4d.ll
index c022a43347..f8d523f2c7 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_5d.ll
index a5a9cef492..f1104484c8 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_1d.ll
index 2e2b3a86bc..d25cd2aa46 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_2d.ll
index b2800ffc98..c2fdb514b6 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_3d.ll
index 0a667ceaf5..a77f118b64 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_4d.ll
index 14711c751c..a171fba27f 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_5d.ll
index 0f28a56460..1538221ea6 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_min_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_3d.ll
index b163c49caa..d3832a0185 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_4d.ll
index 375b37477d..9d7a79214c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_5d.ll
index 64cd8a78b1..2b8f449c46 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_1d.ll
index d1471d795e..ea00c76c5b 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_2d.ll
index f5e8cc6daf..b259dd25a2 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_3d.ll
index 0dfcb9d37c..d485f438b6 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_4d.ll
index d703ae7903..2f33dd8e47 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_5d.ll
index 6f7c1d709b..6238356da1 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_or_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_3d.ll
index dc8eb106e1..a9cf6db63d 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_4d.ll
index 9c41388b19..d14f710cef 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_5d.ll
index 60a9c2278a..299203b666 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_im2col_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_1d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_1d.ll
index 1fa891fd77..8d20bef888 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_1d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_1d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_2d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_2d.ll
index 5474511ebd..e3ea390a0b 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_2d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_2d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_3d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_3d.ll
index b8b0fae3a1..7c785b4521 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_3d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_3d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_4d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_4d.ll
index 840ca4b4dd..5f0fadf127 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_4d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_4d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_5d.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_5d.ll
index 406c297013..cfa609bfc4 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_5d.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_tensor_reduce_xor_tile_5d.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group.ll
index 9b7fcc408a..d107ebada2 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group_read.ll b/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group_read.ll
index 003374d612..9a8167a5fa 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group_read.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_bulk_wait_group_read.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_ca_16.ll b/cuda-oxide/intrinsics/probes/cp_async_ca_16.ll
index 6b878036c1..fd68791d7b 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_ca_16.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_ca_16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_ca_4.ll b/cuda-oxide/intrinsics/probes/cp_async_ca_4.ll
index d95bfc6ae6..4d7856699e 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_ca_4.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_ca_4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_ca_8.ll b/cuda-oxide/intrinsics/probes/cp_async_ca_8.ll
index 74320807a5..e002dcbaf2 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_ca_8.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_ca_8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_16.ll b/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_16.ll
index 4d7bd1fe1a..c7dd69139f 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_16.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_4.ll b/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_4.ll
index 3639ac6a73..73ea5503d5 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_4.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_8.ll b/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_8.ll
index 0bd93f1922..378d524a1d 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_8.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_ca_zfill_8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_cg_16.ll b/cuda-oxide/intrinsics/probes/cp_async_cg_16.ll
index 1e716fa3d8..9ed6a52162 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_cg_16.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_cg_16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_cg_zfill_16.ll b/cuda-oxide/intrinsics/probes/cp_async_cg_zfill_16.ll
index 31a9f08401..e285d27077 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_cg_zfill_16.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_cg_zfill_16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_commit_group.ll b/cuda-oxide/intrinsics/probes/cp_async_commit_group.ll
index 170b62c9e1..0f8eb28414 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_commit_group.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_commit_group.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive.ll b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive.ll
index a2275af196..0addb10c1a 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc.ll b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc.ll
index 23b21f6087..61f4b8c68b 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc_shared.ll b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc_shared.ll
index 0060317dd7..99de175278 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc_shared.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_noinc_shared.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_shared.ll b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_shared.ll
index e2a4eca4a0..d49656c4e0 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_shared.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_mbarrier_arrive_shared.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_wait_all.ll b/cuda-oxide/intrinsics/probes/cp_async_wait_all.ll
index 9e65e90dd1..26ce456bc6 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_wait_all.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_wait_all.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cp_async_wait_group.ll b/cuda-oxide/intrinsics/probes/cp_async_wait_group.ll
index d47231be2d..347c99ea5c 100644
--- a/cuda-oxide/intrinsics/probes/cp_async_wait_group.ll
+++ b/cuda-oxide/intrinsics/probes/cp_async_wait_group.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_f16x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_f16x2_f32.ll
index 0674c78705..e68eafcffa 100644
--- a/cuda-oxide/intrinsics/probes/cvt_f16x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_f16x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_f32x2_bf16x2.ll b/cuda-oxide/intrinsics/probes/cvt_f32x2_bf16x2.ll
index 70276d8a43..aa9ccb605e 100644
--- a/cuda-oxide/intrinsics/probes/cvt_f32x2_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_f32x2_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e4m3x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e4m3x2.ll
index 63e4f7b437..5feca63319 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e4m3x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e4m3x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e5m2x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e5m2x2.ll
index 30185b23b0..daab8d78d9 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e5m2x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_f16x2_e5m2x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_relu_bf16x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_relu_bf16x2_f32.ll
index bf5a1ce5c2..9e8688dea8 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_relu_bf16x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_relu_bf16x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e4m3x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e4m3x2.ll
index 82db013f53..920123763b 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e4m3x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e4m3x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e5m2x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e5m2x2.ll
index fed9d0a3f7..6e4b6697e5 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e5m2x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_e5m2x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_f32.ll
index 7807de0859..2c100257bb 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_relu_f16x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_relu_satfinite_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_relu_satfinite_tf32_f32.ll
index 03dde0e7eb..72f2068054 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_relu_satfinite_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_relu_satfinite_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_relu_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_relu_tf32_f32.ll
index ec79a85d5b..798279c8b3 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_relu_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_relu_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f16x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f16x2.ll
index 001698b45e..d90d9e288a 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f32.ll
index d77da8e78a..2df0cc6f42 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e4m3x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f16x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f16x2.ll
index a771cfdd27..fc62e942db 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f32.ll
index d40e5056b0..694cc51b18 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_e5m2x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f16x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f16x2.ll
index 38d5553e4f..1df9b892fc 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f32.ll
index f9078aba54..45cdaeb7a7 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e4m3x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f16x2.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f16x2.ll
index 65f3756f9c..b006511c9a 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f32.ll
index 0353b4ee4e..cac2a150ff 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_relu_e5m2x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_tf32_f32.ll
index a345d26bf9..fa711fd95f 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_satfinite_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rn_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rn_tf32_f32.ll
index 7298293841..a1ddc82b97 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rn_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rn_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rna_satfinite_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rna_satfinite_tf32_f32.ll
index b14cf39b8c..764d2a6e71 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rna_satfinite_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rna_satfinite_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rna_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rna_tf32_f32.ll
index ce19a950e7..a5c282a4c9 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rna_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rna_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rz_bf16x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rz_bf16x2_f32.ll
index 685083190f..6a946aed71 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rz_bf16x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rz_bf16x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rz_f16x2_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rz_f16x2_f32.ll
index a4fb07c467..000b7054c0 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rz_f16x2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rz_f16x2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rz_relu_satfinite_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rz_relu_satfinite_tf32_f32.ll
index 95213c1296..f7a12ec46c 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rz_relu_satfinite_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rz_relu_satfinite_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rz_relu_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rz_relu_tf32_f32.ll
index 94080ccd34..733dda6360 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rz_relu_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rz_relu_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rz_satfinite_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rz_satfinite_tf32_f32.ll
index 09cf6b5803..5b6c24128c 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rz_satfinite_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rz_satfinite_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/cvt_rz_tf32_f32.ll b/cuda-oxide/intrinsics/probes/cvt_rz_tf32_f32.ll
index a3a8616f32..f8f39abb08 100644
--- a/cuda-oxide/intrinsics/probes/cvt_rz_tf32_f32.ll
+++ b/cuda-oxide/intrinsics/probes/cvt_rz_tf32_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rm_f32.ll b/cuda-oxide/intrinsics/probes/div_rm_f32.ll
index 0f61c9350e..e8f91ad578 100644
--- a/cuda-oxide/intrinsics/probes/div_rm_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rm_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rm_f64.ll b/cuda-oxide/intrinsics/probes/div_rm_f64.ll
index a190589ea7..09f00888e4 100644
--- a/cuda-oxide/intrinsics/probes/div_rm_f64.ll
+++ b/cuda-oxide/intrinsics/probes/div_rm_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rm_ftz_f32.ll b/cuda-oxide/intrinsics/probes/div_rm_ftz_f32.ll
index 232079de93..0ea9c58769 100644
--- a/cuda-oxide/intrinsics/probes/div_rm_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rm_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rn_f32.ll b/cuda-oxide/intrinsics/probes/div_rn_f32.ll
index 714ec0e096..228cc0aa87 100644
--- a/cuda-oxide/intrinsics/probes/div_rn_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rn_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rn_f64.ll b/cuda-oxide/intrinsics/probes/div_rn_f64.ll
index 4d0717b270..bd90372579 100644
--- a/cuda-oxide/intrinsics/probes/div_rn_f64.ll
+++ b/cuda-oxide/intrinsics/probes/div_rn_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rn_ftz_f32.ll b/cuda-oxide/intrinsics/probes/div_rn_ftz_f32.ll
index b93da31ab7..985d40669e 100644
--- a/cuda-oxide/intrinsics/probes/div_rn_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rn_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rp_f32.ll b/cuda-oxide/intrinsics/probes/div_rp_f32.ll
index c02cc35d92..264c51acac 100644
--- a/cuda-oxide/intrinsics/probes/div_rp_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rp_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rp_f64.ll b/cuda-oxide/intrinsics/probes/div_rp_f64.ll
index a886b936c6..2a63516b60 100644
--- a/cuda-oxide/intrinsics/probes/div_rp_f64.ll
+++ b/cuda-oxide/intrinsics/probes/div_rp_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rp_ftz_f32.ll b/cuda-oxide/intrinsics/probes/div_rp_ftz_f32.ll
index 2505944120..2afa9efd1d 100644
--- a/cuda-oxide/intrinsics/probes/div_rp_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rp_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rz_f32.ll b/cuda-oxide/intrinsics/probes/div_rz_f32.ll
index 586c0d30b9..e0e3aa902a 100644
--- a/cuda-oxide/intrinsics/probes/div_rz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rz_f64.ll b/cuda-oxide/intrinsics/probes/div_rz_f64.ll
index afa12610f9..a0efcae75c 100644
--- a/cuda-oxide/intrinsics/probes/div_rz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/div_rz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/div_rz_ftz_f32.ll b/cuda-oxide/intrinsics/probes/div_rz_ftz_f32.ll
index 6f4bb9b7c5..656c119c87 100644
--- a/cuda-oxide/intrinsics/probes/div_rz_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/div_rz_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dp2a_s32.ll b/cuda-oxide/intrinsics/probes/dp2a_s32.ll
index 3bec0f096c..c876a0c3b8 100644
--- a/cuda-oxide/intrinsics/probes/dp2a_s32.ll
+++ b/cuda-oxide/intrinsics/probes/dp2a_s32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dp2a_u32.ll b/cuda-oxide/intrinsics/probes/dp2a_u32.ll
index 620f3cb82b..f7c3e2f850 100644
--- a/cuda-oxide/intrinsics/probes/dp2a_u32.ll
+++ b/cuda-oxide/intrinsics/probes/dp2a_u32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dp4a_s32.ll b/cuda-oxide/intrinsics/probes/dp4a_s32.ll
index 59e3002b8a..ac0329ea6f 100644
--- a/cuda-oxide/intrinsics/probes/dp4a_s32.ll
+++ b/cuda-oxide/intrinsics/probes/dp4a_s32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dp4a_u32.ll b/cuda-oxide/intrinsics/probes/dp4a_u32.ll
index 28e9ba0263..9a6bbb84c9 100644
--- a/cuda-oxide/intrinsics/probes/dp4a_u32.ll
+++ b/cuda-oxide/intrinsics/probes/dp4a_u32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dsmem_read_u32.ll b/cuda-oxide/intrinsics/probes/dsmem_read_u32.ll
index 1af51dc3da..693c323247 100644
--- a/cuda-oxide/intrinsics/probes/dsmem_read_u32.ll
+++ b/cuda-oxide/intrinsics/probes/dsmem_read_u32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dynamic_smem_size.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/dynamic_smem_size.lib_nvvm.ll
index fd2a06a984..89000848ba 100644
--- a/cuda-oxide/intrinsics/probes/dynamic_smem_size.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/dynamic_smem_size.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/dynamic_smem_size.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/dynamic_smem_size.llvm_nvptx.ll
index dd1a391113..b0ae0bf16e 100644
--- a/cuda-oxide/intrinsics/probes/dynamic_smem_size.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/dynamic_smem_size.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/elect_sync.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/elect_sync.lib_nvvm.ll
index 62e82e1c05..3ceb7bacc1 100644
--- a/cuda-oxide/intrinsics/probes/elect_sync.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/elect_sync.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/elect_sync.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/elect_sync.llvm_nvptx.ll
index 159f7f611e..56c243c066 100644
--- a/cuda-oxide/intrinsics/probes/elect_sync.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/elect_sync.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/envreg1.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/envreg1.lib_nvvm.ll
index ab2c0faa35..2084941712 100644
--- a/cuda-oxide/intrinsics/probes/envreg1.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/envreg1.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/envreg1.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/envreg1.llvm_nvptx.ll
index 5a34008805..c26aa21244 100644
--- a/cuda-oxide/intrinsics/probes/envreg1.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/envreg1.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/envreg2.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/envreg2.lib_nvvm.ll
index a51f98d74c..57c4d39162 100644
--- a/cuda-oxide/intrinsics/probes/envreg2.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/envreg2.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/envreg2.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/envreg2.llvm_nvptx.ll
index 63942849b8..af884de851 100644
--- a/cuda-oxide/intrinsics/probes/envreg2.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/envreg2.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ex2_approx_f16.ll b/cuda-oxide/intrinsics/probes/ex2_approx_f16.ll
index e72dcc6a2c..dc1d7a4a5b 100644
--- a/cuda-oxide/intrinsics/probes/ex2_approx_f16.ll
+++ b/cuda-oxide/intrinsics/probes/ex2_approx_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ex2_approx_f32.ll b/cuda-oxide/intrinsics/probes/ex2_approx_f32.ll
index d9ca0c7e96..b0d7997a38 100644
--- a/cuda-oxide/intrinsics/probes/ex2_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/ex2_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ex2_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/ex2_approx_ftz_f32.ll
index 5b3c4a6c53..7306e10b93 100644
--- a/cuda-oxide/intrinsics/probes/ex2_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/ex2_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_mbarrier_init_release_cluster.ll b/cuda-oxide/intrinsics/probes/fence_mbarrier_init_release_cluster.ll
index 709ea6786b..ad98279bac 100644
--- a/cuda-oxide/intrinsics/probes/fence_mbarrier_init_release_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/fence_mbarrier_init_release_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_acquire_shared_cluster_cluster.ll b/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_acquire_shared_cluster_cluster.ll
index d37dcb85c4..54f2db264f 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_acquire_shared_cluster_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_acquire_shared_cluster_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_release_shared_cta_cluster.ll b/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_release_shared_cta_cluster.ll
index 0420e795fb..fd539cbdcd 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_release_shared_cta_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_async_generic_release_shared_cta_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_async_shared_cta.ll b/cuda-oxide/intrinsics/probes/fence_proxy_async_shared_cta.ll
index 756ab152ef..07cdb52a2a 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_async_shared_cta.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_async_shared_cta.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cluster.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cluster.ll
index 9795cc8a82..1af7bc56ea 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cta.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cta.ll
index af28e9032a..a1e3f909a1 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cta.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_cta.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_gpu.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_gpu.ll
index e01e367782..b2b7da6b0a 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_gpu.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_gpu.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_system.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_system.ll
index 99925c3ec9..d02c4efd94 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_system.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_acquire_system.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cluster.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cluster.ll
index a01a1e6b9a..d63f549758 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cta.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cta.ll
index 435e4e97c8..dbdb54654f 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cta.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_cta.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_gpu.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_gpu.ll
index 15ab9ca536..c63b505f08 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_gpu.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_gpu.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_system.ll b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_system.ll
index 61f2340f44..a48d2cb019 100644
--- a/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_system.ll
+++ b/cuda-oxide/intrinsics/probes/fence_proxy_tensormap_generic_release_system.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_bf16x2.ll b/cuda-oxide/intrinsics/probes/fma_bf16x2.ll
index d03cb35b49..0227a87cf5 100644
--- a/cuda-oxide/intrinsics/probes/fma_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_f16x2.ll b/cuda-oxide/intrinsics/probes/fma_f16x2.ll
index 11570cb7dc..3006ca5867 100644
--- a/cuda-oxide/intrinsics/probes/fma_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_f32x2.ll b/cuda-oxide/intrinsics/probes/fma_f32x2.ll
index 1441f8cb69..4178195aff 100644
--- a/cuda-oxide/intrinsics/probes/fma_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_ftz_f16x2.ll b/cuda-oxide/intrinsics/probes/fma_ftz_f16x2.ll
index 56787d210a..1dab638241 100644
--- a/cuda-oxide/intrinsics/probes/fma_ftz_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_ftz_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_ftz_f32x2.ll b/cuda-oxide/intrinsics/probes/fma_ftz_f32x2.ll
index aea071a677..88f236546c 100644
--- a/cuda-oxide/intrinsics/probes/fma_ftz_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_ftz_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_ftz_relu_f16x2.ll b/cuda-oxide/intrinsics/probes/fma_ftz_relu_f16x2.ll
index f95f66a01f..e6ce5730d2 100644
--- a/cuda-oxide/intrinsics/probes/fma_ftz_relu_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_ftz_relu_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_ftz_sat_f16x2.ll b/cuda-oxide/intrinsics/probes/fma_ftz_sat_f16x2.ll
index 3b91ebee55..69c4856a3f 100644
--- a/cuda-oxide/intrinsics/probes/fma_ftz_sat_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_ftz_sat_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_relu_bf16x2.ll b/cuda-oxide/intrinsics/probes/fma_relu_bf16x2.ll
index b9782e147a..1f06311c75 100644
--- a/cuda-oxide/intrinsics/probes/fma_relu_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_relu_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_relu_f16x2.ll b/cuda-oxide/intrinsics/probes/fma_relu_f16x2.ll
index 5ce2a6ce6a..85fc23c5e9 100644
--- a/cuda-oxide/intrinsics/probes/fma_relu_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_relu_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rm_f32.ll b/cuda-oxide/intrinsics/probes/fma_rm_f32.ll
index f0c4c72614..bd78e7fec2 100644
--- a/cuda-oxide/intrinsics/probes/fma_rm_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rm_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rm_f64.ll b/cuda-oxide/intrinsics/probes/fma_rm_f64.ll
index cfc8075316..172f5a198f 100644
--- a/cuda-oxide/intrinsics/probes/fma_rm_f64.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rm_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rm_ftz_f32.ll b/cuda-oxide/intrinsics/probes/fma_rm_ftz_f32.ll
index 6324e54129..15c7ff5a84 100644
--- a/cuda-oxide/intrinsics/probes/fma_rm_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rm_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rm_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rm_ftz_sat_f32.ll
index f4bf2b34c0..d5d90f504e 100644
--- a/cuda-oxide/intrinsics/probes/fma_rm_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rm_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rm_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rm_sat_f32.ll
index 7adc765483..f6a0afb978 100644
--- a/cuda-oxide/intrinsics/probes/fma_rm_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rm_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rn_f32.ll b/cuda-oxide/intrinsics/probes/fma_rn_f32.ll
index 880d1833af..fe95c6605a 100644
--- a/cuda-oxide/intrinsics/probes/fma_rn_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rn_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rn_f64.ll b/cuda-oxide/intrinsics/probes/fma_rn_f64.ll
index 7a21f94476..aa8de7458c 100644
--- a/cuda-oxide/intrinsics/probes/fma_rn_f64.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rn_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rn_ftz_f32.ll b/cuda-oxide/intrinsics/probes/fma_rn_ftz_f32.ll
index 5701f6a574..7f8f6bcdb5 100644
--- a/cuda-oxide/intrinsics/probes/fma_rn_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rn_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rn_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rn_ftz_sat_f32.ll
index c4eb49e0fe..7675704ac7 100644
--- a/cuda-oxide/intrinsics/probes/fma_rn_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rn_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rn_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rn_sat_f32.ll
index 9ecae992d5..abd1aa265f 100644
--- a/cuda-oxide/intrinsics/probes/fma_rn_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rn_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rp_f32.ll b/cuda-oxide/intrinsics/probes/fma_rp_f32.ll
index 5c7e2e252b..c2c322df2a 100644
--- a/cuda-oxide/intrinsics/probes/fma_rp_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rp_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rp_f64.ll b/cuda-oxide/intrinsics/probes/fma_rp_f64.ll
index a69a5ffb03..64bc22b5ab 100644
--- a/cuda-oxide/intrinsics/probes/fma_rp_f64.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rp_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rp_ftz_f32.ll b/cuda-oxide/intrinsics/probes/fma_rp_ftz_f32.ll
index 59c486b424..e938739361 100644
--- a/cuda-oxide/intrinsics/probes/fma_rp_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rp_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rp_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rp_ftz_sat_f32.ll
index eb3e059038..39143c43a8 100644
--- a/cuda-oxide/intrinsics/probes/fma_rp_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rp_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rp_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rp_sat_f32.ll
index cd7e49669a..795d4c65bd 100644
--- a/cuda-oxide/intrinsics/probes/fma_rp_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rp_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rz_f32.ll b/cuda-oxide/intrinsics/probes/fma_rz_f32.ll
index fad52e7945..4d0f422f23 100644
--- a/cuda-oxide/intrinsics/probes/fma_rz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rz_f64.ll b/cuda-oxide/intrinsics/probes/fma_rz_f64.ll
index b26219dbb9..22b01235a1 100644
--- a/cuda-oxide/intrinsics/probes/fma_rz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rz_ftz_f32.ll b/cuda-oxide/intrinsics/probes/fma_rz_ftz_f32.ll
index 0f7d190873..ed8b184bc3 100644
--- a/cuda-oxide/intrinsics/probes/fma_rz_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rz_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rz_ftz_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rz_ftz_sat_f32.ll
index 774bdb33fc..03b8c41659 100644
--- a/cuda-oxide/intrinsics/probes/fma_rz_ftz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rz_ftz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_rz_sat_f32.ll b/cuda-oxide/intrinsics/probes/fma_rz_sat_f32.ll
index e8b2396b07..ef0b109944 100644
--- a/cuda-oxide/intrinsics/probes/fma_rz_sat_f32.ll
+++ b/cuda-oxide/intrinsics/probes/fma_rz_sat_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/fma_sat_f16x2.ll b/cuda-oxide/intrinsics/probes/fma_sat_f16x2.ll
index c69f4d0b0d..bb61385a0f 100644
--- a/cuda-oxide/intrinsics/probes/fma_sat_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/fma_sat_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/globaltimer.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/globaltimer.lib_nvvm.ll
index bd0606ec80..341ecde82b 100644
--- a/cuda-oxide/intrinsics/probes/globaltimer.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/globaltimer.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/globaltimer.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/globaltimer.llvm_nvptx.ll
index f8833f42c6..c562b17969 100644
--- a/cuda-oxide/intrinsics/probes/globaltimer.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/globaltimer.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/grid_dependency_launch_dependents.ll b/cuda-oxide/intrinsics/probes/grid_dependency_launch_dependents.ll
index 8793515f3e..b90c665711 100644
--- a/cuda-oxide/intrinsics/probes/grid_dependency_launch_dependents.ll
+++ b/cuda-oxide/intrinsics/probes/grid_dependency_launch_dependents.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/grid_dependency_wait.ll b/cuda-oxide/intrinsics/probes/grid_dependency_wait.ll
index 3dcc697fda..2cbb8fa061 100644
--- a/cuda-oxide/intrinsics/probes/grid_dependency_wait.ll
+++ b/cuda-oxide/intrinsics/probes/grid_dependency_wait.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/grid_dim_x.ll b/cuda-oxide/intrinsics/probes/grid_dim_x.ll
index 6160b58479..d46c40c165 100644
--- a/cuda-oxide/intrinsics/probes/grid_dim_x.ll
+++ b/cuda-oxide/intrinsics/probes/grid_dim_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/grid_dim_y.ll b/cuda-oxide/intrinsics/probes/grid_dim_y.ll
index e568eef09c..00341c087b 100644
--- a/cuda-oxide/intrinsics/probes/grid_dim_y.ll
+++ b/cuda-oxide/intrinsics/probes/grid_dim_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/grid_dim_z.ll b/cuda-oxide/intrinsics/probes/grid_dim_z.ll
index e3ab0c2c78..7cfe56e7bc 100644
--- a/cuda-oxide/intrinsics/probes/grid_dim_z.ll
+++ b/cuda-oxide/intrinsics/probes/grid_dim_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/gridid.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/gridid.lib_nvvm.ll
index 91bf8bae59..a2b3c2796c 100644
--- a/cuda-oxide/intrinsics/probes/gridid.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/gridid.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/gridid.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/gridid.llvm_nvptx.ll
index 4d3af9c030..4bf2bb73f3 100644
--- a/cuda-oxide/intrinsics/probes/gridid.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/gridid.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lane_id.ll b/cuda-oxide/intrinsics/probes/lane_id.ll
index 4375748889..3810a36af7 100644
--- a/cuda-oxide/intrinsics/probes/lane_id.ll
+++ b/cuda-oxide/intrinsics/probes/lane_id.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lanemask_eq.ll b/cuda-oxide/intrinsics/probes/lanemask_eq.ll
index 3f0c5becda..20799aed61 100644
--- a/cuda-oxide/intrinsics/probes/lanemask_eq.ll
+++ b/cuda-oxide/intrinsics/probes/lanemask_eq.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lanemask_ge.ll b/cuda-oxide/intrinsics/probes/lanemask_ge.ll
index dcaf2c81f7..f29b87e61f 100644
--- a/cuda-oxide/intrinsics/probes/lanemask_ge.ll
+++ b/cuda-oxide/intrinsics/probes/lanemask_ge.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lanemask_gt.ll b/cuda-oxide/intrinsics/probes/lanemask_gt.ll
index 4d5ea2baab..53a7542517 100644
--- a/cuda-oxide/intrinsics/probes/lanemask_gt.ll
+++ b/cuda-oxide/intrinsics/probes/lanemask_gt.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lanemask_le.ll b/cuda-oxide/intrinsics/probes/lanemask_le.ll
index 42a538ad53..924858d48a 100644
--- a/cuda-oxide/intrinsics/probes/lanemask_le.ll
+++ b/cuda-oxide/intrinsics/probes/lanemask_le.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lanemask_lt.ll b/cuda-oxide/intrinsics/probes/lanemask_lt.ll
index 29445102d1..189d4d8182 100644
--- a/cuda-oxide/intrinsics/probes/lanemask_lt.ll
+++ b/cuda-oxide/intrinsics/probes/lanemask_lt.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8.ll
index 48aae7f31f..2ba6ad2898 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64.ll
index 335ea7f964..6056d52646 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32.ll
index 46110462c5..7eeffa64bd 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x1_trans_b8x16_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8.ll
index e54bd8ddb6..c64347b04d 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64.ll
index 186e16a9f4..0333251475 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32.ll
index 48a46dd253..5f2ccf3356 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m16n16_x2_trans_b8x16_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b4x16_p64.ll
index a14091597c..1819f1a48a 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b6x16_p32.ll
index 300fe10d75..fb06db92da 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x1_b8x16_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b4x16_p64.ll
index 4d08f3a647..d610e0ceba 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b6x16_p32.ll
index 04357788fb..4ddddf4817 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x2_b8x16_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b4x16_p64.ll
index 47a24b0408..e6ea8ec6e2 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b6x16_p32.ll
index 16b6279264..216a433048 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n16_x4_b8x16_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_b16.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_b16.ll
index 6de5101df0..3f3b32a4b7 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_b16.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_trans_b16.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_trans_b16.ll
index 2e62e6f8bc..4c4f562852 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_trans_b16.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x1_trans_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_b16.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_b16.ll
index ff9314f5c5..5437865344 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_b16.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_trans_b16.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_trans_b16.ll
index c8a8192730..38c1e31d97 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_trans_b16.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x2_trans_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_b16.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_b16.ll
index ef545a1dec..f823d1f56d 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_b16.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_trans_b16.ll b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_trans_b16.ll
index 8feaa576cf..2c7a254706 100644
--- a/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_trans_b16.ll
+++ b/cuda-oxide/intrinsics/probes/ldmatrix_m8n8_x4_trans_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lg2_approx_f32.ll b/cuda-oxide/intrinsics/probes/lg2_approx_f32.ll
index f162fddc72..d20f9b8100 100644
--- a/cuda-oxide/intrinsics/probes/lg2_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/lg2_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/lg2_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/lg2_approx_ftz_f32.ll
index fe37b75dd4..3150cdcf51 100644
--- a/cuda-oxide/intrinsics/probes/lg2_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/lg2_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/map_shared_rank.ll b/cuda-oxide/intrinsics/probes/map_shared_rank.ll
index 3e92103cf9..17ee6027fd 100644
--- a/cuda-oxide/intrinsics/probes/map_shared_rank.ll
+++ b/cuda-oxide/intrinsics/probes/map_shared_rank.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/match_all_i64_sync.ll b/cuda-oxide/intrinsics/probes/match_all_i64_sync.ll
index 5c4808bd9a..c0c2733487 100644
--- a/cuda-oxide/intrinsics/probes/match_all_i64_sync.ll
+++ b/cuda-oxide/intrinsics/probes/match_all_i64_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/match_all_sync.ll b/cuda-oxide/intrinsics/probes/match_all_sync.ll
index 10c5ee442b..b076a21e01 100644
--- a/cuda-oxide/intrinsics/probes/match_all_sync.ll
+++ b/cuda-oxide/intrinsics/probes/match_all_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/match_any_i64_sync.ll b/cuda-oxide/intrinsics/probes/match_any_i64_sync.ll
index 90b70fd545..cfc9acc3e1 100644
--- a/cuda-oxide/intrinsics/probes/match_any_i64_sync.ll
+++ b/cuda-oxide/intrinsics/probes/match_any_i64_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/match_any_sync.ll b/cuda-oxide/intrinsics/probes/match_any_sync.ll
index 9288ae011d..e4e65e322a 100644
--- a/cuda-oxide/intrinsics/probes/match_any_sync.ll
+++ b/cuda-oxide/intrinsics/probes/match_any_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_bf16.ll b/cuda-oxide/intrinsics/probes/max_bf16.ll
index 42e2be579f..77be96d793 100644
--- a/cuda-oxide/intrinsics/probes/max_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/max_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_bf16x2.ll b/cuda-oxide/intrinsics/probes/max_bf16x2.ll
index ce7f8125b1..7150683ad3 100644
--- a/cuda-oxide/intrinsics/probes/max_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_f16.ll b/cuda-oxide/intrinsics/probes/max_f16.ll
index e208b199e1..d12d32818b 100644
--- a/cuda-oxide/intrinsics/probes/max_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_f16x2.ll b/cuda-oxide/intrinsics/probes/max_f16x2.ll
index 8789a5d3ec..eb099b1f93 100644
--- a/cuda-oxide/intrinsics/probes/max_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_f16.ll b/cuda-oxide/intrinsics/probes/max_ftz_f16.ll
index 458003c675..9ce110e2e6 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_f16x2.ll b/cuda-oxide/intrinsics/probes/max_ftz_f16x2.ll
index 1bd819eaa1..99431d33f4 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_nan_f16.ll b/cuda-oxide/intrinsics/probes/max_ftz_nan_f16.ll
index f3ed7912a0..5b041b2ded 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_nan_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_nan_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_nan_f16x2.ll b/cuda-oxide/intrinsics/probes/max_ftz_nan_f16x2.ll
index cfc8fec1dd..3fce4e7b67 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_nan_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_nan_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16.ll
index dcc8bf5ce5..5ccb70e93a 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16x2.ll
index e1a1ac6d9c..d1b271ce91 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f32.ll
index 5aab79c28f..0663c3c250 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_nan_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16.ll
index 06a882d0cd..6c218589e8 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16x2.ll
index 5cee3d6a52..289a6f955e 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f32.ll
index f6fed3626a..3fa48796ac 100644
--- a/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/max_ftz_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_bf16.ll b/cuda-oxide/intrinsics/probes/max_nan_bf16.ll
index 20f90efb9c..9239b9fe52 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_bf16x2.ll b/cuda-oxide/intrinsics/probes/max_nan_bf16x2.ll
index 4938b05160..0da393f4ff 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_f16.ll b/cuda-oxide/intrinsics/probes/max_nan_f16.ll
index b48856a401..f91129746f 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_f16x2.ll b/cuda-oxide/intrinsics/probes/max_nan_f16x2.ll
index f7840ca365..4656697104 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16.ll b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16.ll
index f234e4ba90..974357c43b 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16x2.ll b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16x2.ll
index 0e1a0e22ed..437daaa6bb 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16.ll
index 7f977d3f5c..5d73b53155 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16x2.ll
index ad91a7b8cd..ab4af618e8 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f32.ll
index 505dc8d7c1..4f558e905c 100644
--- a/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/max_nan_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_relu_s16x2.ll b/cuda-oxide/intrinsics/probes/max_relu_s16x2.ll
index 25a6d49c6b..bd8265fe52 100644
--- a/cuda-oxide/intrinsics/probes/max_relu_s16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_relu_s16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_relu_s32.ll b/cuda-oxide/intrinsics/probes/max_relu_s32.ll
index d364154157..5e8e6ed0d7 100644
--- a/cuda-oxide/intrinsics/probes/max_relu_s32.ll
+++ b/cuda-oxide/intrinsics/probes/max_relu_s32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_s16x2.ll b/cuda-oxide/intrinsics/probes/max_s16x2.ll
index 1860d6088d..db5370ba75 100644
--- a/cuda-oxide/intrinsics/probes/max_s16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_s16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_u16x2.ll b/cuda-oxide/intrinsics/probes/max_u16x2.ll
index d2a27f7c79..6a2e6c31c9 100644
--- a/cuda-oxide/intrinsics/probes/max_u16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_u16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16.ll b/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16.ll
index 174a9ad97b..03e42f432b 100644
--- a/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16x2.ll b/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16x2.ll
index 6c9213a7a4..a73ef96512 100644
--- a/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_xorsign_abs_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16.ll
index 12c7994b29..f053059555 100644
--- a/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16x2.ll
index b7ca356ab1..04aefd53a3 100644
--- a/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/max_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/max_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/max_xorsign_abs_f32.ll
index e0be80518f..ca09146c29 100644
--- a/cuda-oxide/intrinsics/probes/max_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/max_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_arrive.ll b/cuda-oxide/intrinsics/probes/mbarrier_arrive.ll
index b592698137..1b0d581793 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_arrive.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_arrive.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_arrive_cluster.ll b/cuda-oxide/intrinsics/probes/mbarrier_arrive_cluster.ll
index 93bb09121d..aff83d50be 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_arrive_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_arrive_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx.ll b/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx.ll
index c36d6aad5c..4e35deeb04 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx_cluster.ll b/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx_cluster.ll
index ad001c0ebb..65fe02b980 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_arrive_expect_tx_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_arrive_no_complete.ll b/cuda-oxide/intrinsics/probes/mbarrier_arrive_no_complete.ll
index fc17a2816f..3cf7ee744f 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_arrive_no_complete.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_arrive_no_complete.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_init.ll b/cuda-oxide/intrinsics/probes/mbarrier_init.ll
index ab6038402e..910ac349d8 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_init.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_init.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_inval.ll b/cuda-oxide/intrinsics/probes/mbarrier_inval.ll
index 21d816dafa..2c7fae458f 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_inval.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_inval.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_test_wait.ll b/cuda-oxide/intrinsics/probes/mbarrier_test_wait.ll
index e16b8dc05c..0e21dc951d 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_test_wait.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_test_wait.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_try_wait.ll b/cuda-oxide/intrinsics/probes/mbarrier_try_wait.ll
index 9ea2000fdc..df497af274 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_try_wait.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_try_wait.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity.ll b/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity.ll
index ddfe48da54..8374649481 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity_cluster.ll b/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity_cluster.ll
index 609bfd5de7..128985b3ea 100644
--- a/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/mbarrier_try_wait_parity_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_bf16.ll b/cuda-oxide/intrinsics/probes/min_bf16.ll
index 685c8beee2..3bbd2641ca 100644
--- a/cuda-oxide/intrinsics/probes/min_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/min_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_bf16x2.ll b/cuda-oxide/intrinsics/probes/min_bf16x2.ll
index 98d2a848a7..f57634c72f 100644
--- a/cuda-oxide/intrinsics/probes/min_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_f16.ll b/cuda-oxide/intrinsics/probes/min_f16.ll
index db6d4cf788..a944cfb4dc 100644
--- a/cuda-oxide/intrinsics/probes/min_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_f16x2.ll b/cuda-oxide/intrinsics/probes/min_f16x2.ll
index fd6aad9c0a..f295381bb6 100644
--- a/cuda-oxide/intrinsics/probes/min_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_f16.ll b/cuda-oxide/intrinsics/probes/min_ftz_f16.ll
index d7f60d85a2..3d9104e413 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_f16x2.ll b/cuda-oxide/intrinsics/probes/min_ftz_f16x2.ll
index 2c0522f034..b06fa76805 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_nan_f16.ll b/cuda-oxide/intrinsics/probes/min_ftz_nan_f16.ll
index 40383f5720..00d289910f 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_nan_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_nan_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_nan_f16x2.ll b/cuda-oxide/intrinsics/probes/min_ftz_nan_f16x2.ll
index 9d8379e621..b583273e3c 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_nan_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_nan_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16.ll
index b97e669b2c..aa8217ccf9 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16x2.ll
index 984663ea51..53ceb1f1a5 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f32.ll
index 5b1816dc56..037e2349a9 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_nan_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16.ll
index c6cd152c15..153ba0e6d6 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16x2.ll
index 535f2934a3..d2e8fa9463 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f32.ll
index d4b65452e4..27fd6d5ccc 100644
--- a/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/min_ftz_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_bf16.ll b/cuda-oxide/intrinsics/probes/min_nan_bf16.ll
index 800e9c770b..c533c9544b 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_bf16x2.ll b/cuda-oxide/intrinsics/probes/min_nan_bf16x2.ll
index bfc7d311a0..20513a48e4 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_f16.ll b/cuda-oxide/intrinsics/probes/min_nan_f16.ll
index a591854fb3..1a5d56cda6 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_f16x2.ll b/cuda-oxide/intrinsics/probes/min_nan_f16x2.ll
index 4761a5aef5..16d01c1151 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16.ll b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16.ll
index 92b345ae9e..1607ff0343 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16x2.ll b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16x2.ll
index 8d77182d42..e65f385040 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16.ll
index 42ae49f2f1..35b19c4fd1 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16x2.ll
index 4f671739a8..f87889e445 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f32.ll
index 86ecde55d6..653e8bf18d 100644
--- a/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/min_nan_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_relu_s16x2.ll b/cuda-oxide/intrinsics/probes/min_relu_s16x2.ll
index 114ba0ea2b..c08d69fe08 100644
--- a/cuda-oxide/intrinsics/probes/min_relu_s16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_relu_s16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_relu_s32.ll b/cuda-oxide/intrinsics/probes/min_relu_s32.ll
index 72ec4e9862..ab32ac384a 100644
--- a/cuda-oxide/intrinsics/probes/min_relu_s32.ll
+++ b/cuda-oxide/intrinsics/probes/min_relu_s32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_s16x2.ll b/cuda-oxide/intrinsics/probes/min_s16x2.ll
index 024ac9a79d..2f9c14a0d2 100644
--- a/cuda-oxide/intrinsics/probes/min_s16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_s16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_u16x2.ll b/cuda-oxide/intrinsics/probes/min_u16x2.ll
index 17efe05f7d..211614bc5d 100644
--- a/cuda-oxide/intrinsics/probes/min_u16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_u16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16.ll b/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16.ll
index b2c98a1396..c4e6e964e5 100644
--- a/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16x2.ll b/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16x2.ll
index 5c5eda21cc..3b46362412 100644
--- a/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_xorsign_abs_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16.ll b/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16.ll
index fad3ee263c..9e9415875f 100644
--- a/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16.ll
+++ b/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16x2.ll b/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16x2.ll
index 83e2d083f1..0c5a76f1d3 100644
--- a/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/min_xorsign_abs_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/min_xorsign_abs_f32.ll b/cuda-oxide/intrinsics/probes/min_xorsign_abs_f32.ll
index 0600e57238..9486508752 100644
--- a/cuda-oxide/intrinsics/probes/min_xorsign_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/min_xorsign_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_and_popc.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_and_popc.ll
index d963d02bf2..1ea6db4179 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_and_popc.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_and_popc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_xor_popc.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_xor_popc.ll
index ff9b31661d..f343c80ca5 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_xor_popc.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k128_s32_b1_xor_popc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_f16_f16.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_f16_f16.ll
index 61856d680a..f147489e24 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_f16_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_f16_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_bf16.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_bf16.ll
index c7143f7d5c..7776405699 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_f16.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_f16.ll
index 7c10c34523..8e3187734e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_f32_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e4m3.ll
index 68c5760734..2f60560537 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e5m2.ll
index 3787a8eaf1..0107ddc2be 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e4m3.ll
index 7bbb1fb8f0..348ca5a36b 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e5m2.ll
index e93e734e66..044208ada0 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f16_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e4m3.ll
index e4996d459a..da44fee12a 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e5m2.ll
index 0bcfef72ab..cb86d24427 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e4m3.ll
index 83b9d915f4..10859f722d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e5m2.ll
index 9bbd1478a9..0ba1467481 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_fp8_f32_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8.ll
index 9f41f357ca..92879daa3a 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_satfinite.ll
index 0d8a8d7e62..07c76884b2 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8.ll
index 26470d0556..3fa61efac6 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8_satfinite.ll
index 6fa872212a..a5032ec5cb 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8.ll
index 1b7a2ab311..70397939ac 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8.ll
index e1d99bcc83..39b5948174 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8_satfinite.ll
index 561ca82bce..f6f7f63e30 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_satfinite.ll
index b06c35609f..3038cd2516 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k16_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_and_popc.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_and_popc.ll
index 89a369f4e3..e7ed2d289f 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_and_popc.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_and_popc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_xor_popc.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_xor_popc.ll
index 715d24a899..7db1a4bb35 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_xor_popc.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k256_s32_b1_xor_popc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m1.ll
index e46e29201b..4f85a99466 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m3.ll
index b02d641a4e..d2104e7fb4 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e3m2.ll
index 671c967568..8ca6cb55a3 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e4m3.ll
index 16314882c7..db9953f57d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e5m2.ll
index 78fc634a97..87a760c4ff 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m1_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m1.ll
index 60a7744103..dfd197e064 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m3.ll
index 3b21922940..735b58026a 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e3m2.ll
index dcd9c3f56a..eaa10620d8 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e4m3.ll
index 8a15719960..e44cce86e8 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e5m2.ll
index 975bc9ed11..8fbc3419a0 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e2m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m1.ll
index caf1cf19bd..e3f1714a86 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m3.ll
index 97a7ad5840..76c3c32edf 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e3m2.ll
index 04bc4acf53..3c8be80062 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e4m3.ll
index 72f90eed95..eee0fafb8f 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e5m2.ll
index 202ca3ce08..fd52e2f349 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e3m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m1.ll
index 2038d886f5..3baa32de0e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m3.ll
index defcac001d..add7cc4475 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e3m2.ll
index cd73de8eb0..9e7c95d969 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e4m3.ll
index 4431e4d8a5..4b7122373b 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e5m2.ll
index aac88c2777..c4ec2080fb 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m1.ll
index 56d53ae963..630f56bf62 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m3.ll
index 94f34125c8..fecfeab6b8 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e3m2.ll
index cdd7e57398..1a0bc3f45b 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e4m3.ll
index 1ad6123d34..bc46a827c3 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e5m2.ll
index a23db44a2c..951a05b0f6 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f16_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m1.ll
index d13ede0396..96c95b15a5 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m3.ll
index af6ae69ee0..b0b2bcc4a1 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e3m2.ll
index b9fb90a7cf..52db2b582e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e4m3.ll
index b65a5e1edf..84c6daf029 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e5m2.ll
index d8ae5eb98d..022877ae35 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m1_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m1.ll
index f76a320e85..96c70b0d9e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m3.ll
index e186591fc6..97fe74ef02 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e3m2.ll
index 06fb18bdf6..69f1fabc6e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e4m3.ll
index 6330da3521..de067d5495 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e5m2.ll
index 90ac832473..ed9138286c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e2m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m1.ll
index 89a311470b..1c775c4e24 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m3.ll
index fc423843d7..4cc5ed515c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e3m2.ll
index 62adbab13e..6f432aa964 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e4m3.ll
index 3c95f08757..5141ec0786 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e5m2.ll
index 595955a2bf..ffccc1412f 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e3m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m1.ll
index fe76080066..b4d7276108 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m3.ll
index 355be7c4d3..1221b46de6 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e3m2.ll
index 0ed0557252..779b69c37b 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e4m3.ll
index 726e27d404..b51a697a76 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e5m2.ll
index a20d6e3566..767763cb26 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m1.ll
index a05abbbcfe..92c8be40b4 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m3.ll
index e4227f1110..980aa082e0 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e3m2.ll
index f3670b096d..2101d232ac 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e4m3.ll
index f26c6f95fe..46d5d57412 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e5m2.ll
index e97e238d87..4bec7816f6 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_f32_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e4m3.ll
index 02f46cc6bd..c6f7ee644a 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e5m2.ll
index 0704365d94..38c8822a13 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e4m3.ll
index 7fece7f172..823495e8ab 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e5m2.ll
index 76a03fbe61..00aa64998e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f16_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e4m3.ll
index c17965b9e1..fd74bfbf88 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e5m2.ll
index ec293d8e6c..751d6de387 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e4m3.ll
index fcfe25612f..e62f38b979 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e5m2.ll
index d5222ca264..9bb91ff4be 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_fp8_f32_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1.ll
index fcf989dcdb..6ac77efe23 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3.ll
index 497b1c61a8..a29c45e41e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2.ll
index 201c6b4011..8b6865a36d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3.ll
index a781318992..f1a47b1431 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2.ll
index 2dcbc0291f..c141a5ed75 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m1_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1.ll
index 9ab7304639..bcd74f3058 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3.ll
index 6d42709e72..15c81cafb8 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2.ll
index 07fe418d71..1090808214 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3.ll
index 99db7845c6..d8a60527d0 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2.ll
index f0162e7b5b..9a10044a3c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e2m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1.ll
index 8bc029fb6d..c9839d36d0 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3.ll
index c814062020..0ad35af407 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2.ll
index 48ea460094..b3433362fc 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3.ll
index 7177de07ab..d009f0a87d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2.ll
index 2e3f5b56c2..7f33262008 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e3m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1.ll
index 1baca9d509..c3de14683a 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3.ll
index e373189c4d..db4ba8e839 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2.ll
index bb9b4d535d..e220edb8f2 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3.ll
index 9f8922664f..7900e8e9e6 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2.ll
index 83225dc35d..f879eaf258 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e4m3_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1.ll
index b895e7eaf8..7afaf17ef0 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3.ll
index 87a1a68e0a..6b40349352 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2.ll
index 6e971e4550..699d72600e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3.ll
index de42fd30e4..df4a91ccd7 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2.ll
index 4f37af1c1e..7af2a65bb3 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_mxf8f6f4_f32_e5m2_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4.ll
index d98377829a..7b171cbe43 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_satfinite.ll
index 1547583cf9..7eb4d7daae 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4.ll
index 887e048030..c6017a7c15 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4_satfinite.ll
index 8f9a007242..051cc6c569 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8.ll
index 4870c39f1b..b7f1b1db9e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_satfinite.ll
index cfe37b5c89..b206c7932c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8.ll
index 3925250b55..ac519e39fc 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8_satfinite.ll
index 428e6c42e2..d4f89e99bb 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4.ll
index eedaf28c42..209a3d9989 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4.ll
index f5ecd80ef9..a681419efb 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4_satfinite.ll
index 9336a6d1f4..5c96b8193e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_satfinite.ll
index 4dae4a3d74..cecdf1f30f 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8.ll
index d0cd3714f7..7c350cf31b 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8.ll
index f3ca3a052d..0bebe9da87 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8_satfinite.ll
index 7af581e38b..980f0166b2 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_satfinite.ll
index 35145ad43d..0efe91ff83 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k32_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k4_f32_tf32.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k4_f32_tf32.ll
index 1bee661ae6..6e70e7d1ec 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k4_f32_tf32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k4_f32_tf32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4.ll
index d0d6079f1f..e9890ce30d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_satfinite.ll
index b22126cd85..9ee23211c7 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4.ll
index 39a6330ed7..14d8201f7e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4_satfinite.ll
index e098daa90a..19343fb6de 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4.ll
index f408f9bcf9..8826984e26 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4.ll
index c233aa5d2f..3344055106 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4_satfinite.ll
index bfec98968c..a33954d5ee 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_satfinite.ll
index a5e0eb5719..934c12dff5 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k64_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f16_f16.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f16_f16.ll
index d71f481be1..b021c14fac 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f16_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f16_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_bf16.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_bf16.ll
index 0b66faadfd..e64504fff5 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_f16.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_f16.ll
index 2770e8712f..7caff5c91c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_tf32.ll b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_tf32.ll
index 4c3c6fccff..d030101330 100644
--- a/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_tf32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m16n8k8_f32_tf32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_and_popc.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_and_popc.ll
index e66d1165dd..473c2e79de 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_and_popc.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_and_popc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_xor_popc.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_xor_popc.ll
index 7a5cb727fe..3a432cfb15 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_xor_popc.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k128_s32_b1_xor_popc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8.ll
index d6d1b5fa40..adddb22d13 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_satfinite.ll
index feffa7e53e..4a8d18d3da 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8.ll
index 63488fee56..054ea5e279 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8_satfinite.ll
index 72786dd844..9bf0b3be33 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8.ll
index 2a2d346a55..50bc3245d2 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8.ll
index 2c10dac69b..6e4ffb7eb5 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8_satfinite.ll
index fa86ab8e8c..e2160e822c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_satfinite.ll
index 6c6359518c..dd72002d02 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k16_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4.ll
index ab00bf2115..b1d3623e4d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_satfinite.ll
index e7d49c1bef..2ad3a82c38 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4.ll
index 33d8a273c5..530fd956cb 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4_satfinite.ll
index 01ce04f226..53b117b1c6 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4.ll
index ced502c6b4..7b524c3780 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4.ll
index 895b9dfa45..598c261e2d 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4_satfinite.ll
index 9aa33a996e..b69f73456e 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_satfinite.ll
index bf67485621..d3892576bf 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k32_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_m8n8k4_f64.ll b/cuda-oxide/intrinsics/probes/mma_m8n8k4_f64.ll
index dbf88178e5..2e369de46c 100644
--- a/cuda-oxide/intrinsics/probes/mma_m8n8k4_f64.ll
+++ b/cuda-oxide/intrinsics/probes/mma_m8n8k4_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4.ll
index d7f31a77d6..96d692c1ac 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_satfinite.ll
index 41d695ffe7..87cb08ca39 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4.ll
index b6e0ba0453..cfe144b4f5 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4_satfinite.ll
index dd0f05e125..07e3b09849 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4.ll
index 537f839550..54beec52dd 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4.ll
index e5b354e0f4..3ae0c3ab34 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4_satfinite.ll
index a126f1005f..33b8a3e468 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_satfinite.ll
index 3e8feac629..3d7fdf7538 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k128_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8.ll
index d5c6f74a7f..c476d84a31 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_satfinite.ll
index 2f4d672a76..7114cc098e 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8.ll
index a7a9ea84d3..ee3ac32bed 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8_satfinite.ll
index 7f8ea390a9..5cc5b3a4f4 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8.ll
index 9be435913b..97dda1e812 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8.ll
index db618bc374..f05b0a4760 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8_satfinite.ll
index 2916d72356..00a0005305 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_satfinite.ll
index 71f8ad3386..f60dcffd1b 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k32_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e4m3_f32.ll
index 0850246521..7352440cd0 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e5m2_f32.ll
index 099f62af1d..64747e411f 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e4m3_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e4m3_f32.ll
index c3f9e73766..99f94d3258 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e5m2_f32.ll
index 2c966709c8..5cebce211f 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_f32_e5m2_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4.ll
index 964db6ffe5..1a2729aa84 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_satfinite.ll
index e9b7c2712a..4ffcd5c1b4 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4.ll
index 08e9934829..74262a9bd4 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4_satfinite.ll
index 5aafcbe2c4..c798abd261 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8.ll
index b30f068331..f685ca86ab 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_satfinite.ll
index 9da665295b..6d046a2fa9 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8.ll
index 71bb98fcab..b0be097754 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8_satfinite.ll
index d0636ec068..1f9c965c2d 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4.ll
index a4d698320b..4240c1d1fa 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4.ll
index e7dee2e3e9..3011a63159 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4_satfinite.ll
index 483be82f9a..82307523e8 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_satfinite.ll
index a13a90508c..35d18671bf 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8.ll
index 5227f4c146..d70ff12bf7 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8.ll
index 31788fae16..83026e728b 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8_satfinite.ll
index 1f6165d13c..1598d12615 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_satfinite.ll
index 1bf74ed3fb..4332a7521b 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_m16n8k64_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4.ll
index d2c54f1295..684f3a2e70 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite.ll
index e0bcd14872..21bb6d0c6d 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4.ll
index 4ae9f848cb..c54884e450 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite.ll
index 5b5bbe3323..cb2ddfa7aa 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4.ll
index 28870da031..01817551d1 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4.ll
index 3495c366ea..525df8de25 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite.ll
index 6b0a05e818..b8fd3dde7e 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite.ll
index 25707ae659..31643145aa 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k128_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f16_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f16_f16.ll
index 31d5a18c92..4c4533956b 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f16_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f16_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_bf16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_bf16.ll
index f9f7ffa76d..b284e897b7 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_f16.ll
index 5059228b34..f0f1dbd4a4 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_tf32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_tf32.ll
index 12ce6f2a9a..be5dde36bc 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_tf32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k16_f32_tf32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f16_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f16_f16.ll
index bff33f80c7..85056d3498 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f16_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f16_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_bf16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_bf16.ll
index 04ba90da8b..cbba60c4ba 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_f16.ll
index fadc501d15..15c849298e 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_f32_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8.ll
index 1ce1ad0e48..8c42464754 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite.ll
index c86c1ff5bd..b1050defb6 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8.ll
index 439a394053..d88fe3e3eb 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite.ll
index c99c3e53e8..0fd3ba60e1 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8.ll
index f72f045c70..4d1c1830ae 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8.ll
index cf8d239357..b3dab246b3 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite.ll
index 4b2f0007bd..c9e1a911e9 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite.ll
index 65679c06d1..021358fd1d 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k32_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16.ll
index 63e92f2a0e..70b7321527 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m1_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16.ll
index f1857a7789..4914fd926b 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e2m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16.ll
index 667fec5619..6fd7141112 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e3m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16.ll
index 6a51ca130e..6e85ddd82a 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e4m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16.ll
index 2996ed8bfa..02a23a1169 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m1_e5m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16.ll
index 148beeaffe..618904c371 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m1_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16.ll
index 624fe97fe9..2d9b3994b7 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e2m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16.ll
index 77dde641da..456c9a0483 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e3m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16.ll
index 1efd307b51..c5f468014c 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e4m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16.ll
index ebf30bb152..9fc3d9b536 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e2m3_e5m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16.ll
index be386d3014..30466576ba 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m1_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16.ll
index 8d28298011..e56812c604 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e2m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16.ll
index 0a70815a88..83c991c331 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e3m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16.ll
index 12d3322d7d..5ba1ecb190 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e4m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16.ll
index 5972bab465..c3fbfa9f94 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e3m2_e5m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16.ll
index 05bedd4ca7..d32068231a 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m1_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16.ll
index 3c852f3458..da1a18a093 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e2m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16.ll
index ad7aca5d6b..4e2a6b84ec 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e3m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16.ll
index 5651f7057e..60d12b5bba 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e4m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16.ll
index 24b662e07c..a0b26fc7e0 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e4m3_e5m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16.ll
index b59f4b8ae5..7f1f38bee5 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m1_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16.ll
index fc7f857d6f..e51fcbf0e9 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e2m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16.ll
index 2900a5a529..78bffdbec0 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e3m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16.ll
index 3aea58be20..eb254e0845 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e4m3_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16.ll
index ee13db537e..961ec8b463 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f16_e5m2_e5m2_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32.ll
index 417b7daf67..cccadb49d9 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m1_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32.ll
index 15d50059d0..13c28b8e69 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e2m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32.ll
index ce60de9013..87433ab492 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e3m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32.ll
index 383c88c27d..377d250f54 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32.ll
index 4a78e54b7c..4274cd36df 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m1_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32.ll
index 7df3e1f8ec..2f26e87cc5 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m1_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32.ll
index 7d0036c102..78a15d17ca 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e2m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32.ll
index 1858a99fbe..3b5459eae9 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e3m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32.ll
index 2eb2690023..080082879d 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32.ll
index dabdf1f912..2973d38341 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e2m3_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32.ll
index db87bbc8d0..5fd43812e4 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m1_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32.ll
index 6a2747a9f2..eb3ff37c98 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e2m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32.ll
index a520bf42fd..25662838f7 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e3m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32.ll
index d91345bf6d..e4554753f0 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32.ll
index 6a8a68f043..86de360fe1 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e3m2_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32.ll
index badc172fb8..017859b219 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m1_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32.ll
index d3ee678420..e1e2970290 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e2m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32.ll
index 3612d40921..ede1412f6d 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e3m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32.ll
index 30048c40d0..6e33033b3a 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32.ll
index 9d3a909982..ed5694e626 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e4m3_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32.ll
index 4c81d8736c..b034f065e4 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m1_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32.ll
index b921dbc6b4..3f4d201382 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e2m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32.ll
index 6cdbed3935..bca458a7c3 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e3m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32.ll
index a638854bab..3a2cc89340 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e4m3_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32.ll
index 8b7d7805ca..2173d4db76 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_kind_f8f6f4_f32_e5m2_e5m2_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4.ll
index 1e3f034f45..4a9fdedf9a 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite.ll
index 1ebbafd176..20e4d53b8a 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4.ll
index 447bdcf885..ddbe82e11f 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite.ll
index 39880d919a..6aff5c6f88 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s4_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8.ll
index 29e49dfb6a..0421f0ffc6 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite.ll
index e2ca8eaa43..1d59d7a529 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8.ll
index 24e55f62fb..1a9cce88d0 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite.ll
index 7af0218094..17d4263041 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_s8_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4.ll
index 0deadc777a..9c8a310db5 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4.ll
index b444e489a5..491a194d31 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite.ll
index ee14d0a4f8..1d1a5e1c84 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_s4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite.ll
index c79a4f8547..d4e56c6b32 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u4_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8.ll
index 63ce3fa1c1..254cf0452f 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8.ll
index f67d04342e..a03a94023e 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite.ll
index 7450c5e86b..f235abf757 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_s8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite.ll
index c56e48c5f5..0e68928c1e 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k64_s32_u8_satfinite.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k8_f32_tf32.ll b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k8_f32_tf32.ll
index fbbfec4ad9..10b9927d32 100644
--- a/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k8_f32_tf32.ll
+++ b/cuda-oxide/intrinsics/probes/mma_sp_ordered_metadata_m16n8k8_f32_tf32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/movmatrix_trans_b16.ll b/cuda-oxide/intrinsics/probes/movmatrix_trans_b16.ll
index eb61aefd8c..204e98b791 100644
--- a/cuda-oxide/intrinsics/probes/movmatrix_trans_b16.ll
+++ b/cuda-oxide/intrinsics/probes/movmatrix_trans_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_bf16x2.ll b/cuda-oxide/intrinsics/probes/mul_bf16x2.ll
index c1aa9ab7fc..726b7ea310 100644
--- a/cuda-oxide/intrinsics/probes/mul_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/mul_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_f16x2.ll b/cuda-oxide/intrinsics/probes/mul_f16x2.ll
index 54d9358d3f..092afdea4a 100644
--- a/cuda-oxide/intrinsics/probes/mul_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/mul_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_f32x2.ll b/cuda-oxide/intrinsics/probes/mul_f32x2.ll
index 6ba7f1a2cf..7a4daaeeb6 100644
--- a/cuda-oxide/intrinsics/probes/mul_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/mul_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_ftz_f32x2.ll b/cuda-oxide/intrinsics/probes/mul_ftz_f32x2.ll
index 92527a58e0..e0e2b089d4 100644
--- a/cuda-oxide/intrinsics/probes/mul_ftz_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/mul_ftz_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rm_f32.ll b/cuda-oxide/intrinsics/probes/mul_rm_f32.ll
index 2f85919620..1e194d6780 100644
--- a/cuda-oxide/intrinsics/probes/mul_rm_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rm_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rm_f64.ll b/cuda-oxide/intrinsics/probes/mul_rm_f64.ll
index d30395c092..513893e676 100644
--- a/cuda-oxide/intrinsics/probes/mul_rm_f64.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rm_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rm_ftz_f32.ll b/cuda-oxide/intrinsics/probes/mul_rm_ftz_f32.ll
index 1f81ed3f58..86b0f26e72 100644
--- a/cuda-oxide/intrinsics/probes/mul_rm_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rm_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rn_f32.ll b/cuda-oxide/intrinsics/probes/mul_rn_f32.ll
index 10edac5a56..c936466501 100644
--- a/cuda-oxide/intrinsics/probes/mul_rn_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rn_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rn_f64.ll b/cuda-oxide/intrinsics/probes/mul_rn_f64.ll
index 699ba1b364..d6b9062a8a 100644
--- a/cuda-oxide/intrinsics/probes/mul_rn_f64.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rn_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rn_ftz_f32.ll b/cuda-oxide/intrinsics/probes/mul_rn_ftz_f32.ll
index 14b830cd6c..e43e931133 100644
--- a/cuda-oxide/intrinsics/probes/mul_rn_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rn_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rp_f32.ll b/cuda-oxide/intrinsics/probes/mul_rp_f32.ll
index 6d4c8626fa..98b013e2c0 100644
--- a/cuda-oxide/intrinsics/probes/mul_rp_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rp_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rp_f64.ll b/cuda-oxide/intrinsics/probes/mul_rp_f64.ll
index 01943608a3..a0b25e06c2 100644
--- a/cuda-oxide/intrinsics/probes/mul_rp_f64.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rp_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rp_ftz_f32.ll b/cuda-oxide/intrinsics/probes/mul_rp_ftz_f32.ll
index 88058a3757..b5932f240f 100644
--- a/cuda-oxide/intrinsics/probes/mul_rp_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rp_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rz_f32.ll b/cuda-oxide/intrinsics/probes/mul_rz_f32.ll
index 7094191e33..1edbbaaef0 100644
--- a/cuda-oxide/intrinsics/probes/mul_rz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rz_f64.ll b/cuda-oxide/intrinsics/probes/mul_rz_f64.ll
index e4c450ed82..a8f00b6bd3 100644
--- a/cuda-oxide/intrinsics/probes/mul_rz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/mul_rz_ftz_f32.ll b/cuda-oxide/intrinsics/probes/mul_rz_ftz_f32.ll
index 80f9921d36..9fa2605615 100644
--- a/cuda-oxide/intrinsics/probes/mul_rz_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/mul_rz_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/nanosleep.ll b/cuda-oxide/intrinsics/probes/nanosleep.ll
index 0e1169bca3..ef3c013ae5 100644
--- a/cuda-oxide/intrinsics/probes/nanosleep.ll
+++ b/cuda-oxide/intrinsics/probes/nanosleep.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/neg_bf16x2.ll b/cuda-oxide/intrinsics/probes/neg_bf16x2.ll
index c1b3631c78..dc2600aad3 100644
--- a/cuda-oxide/intrinsics/probes/neg_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/neg_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/neg_f16x2.ll b/cuda-oxide/intrinsics/probes/neg_f16x2.ll
index f80d43a2cc..ee5840d2d9 100644
--- a/cuda-oxide/intrinsics/probes/neg_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/neg_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/nsmid.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/nsmid.lib_nvvm.ll
index 7721ebafad..7d0749f632 100644
--- a/cuda-oxide/intrinsics/probes/nsmid.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/nsmid.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/nsmid.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/nsmid.llvm_nvptx.ll
index 4e45521e75..69cffffa8a 100644
--- a/cuda-oxide/intrinsics/probes/nsmid.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/nsmid.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/nwarpid.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/nwarpid.lib_nvvm.ll
index 3e3bb08f14..b9f83e5cb8 100644
--- a/cuda-oxide/intrinsics/probes/nwarpid.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/nwarpid.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/nwarpid.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/nwarpid.llvm_nvptx.ll
index 05c381ad9c..ee1c99f101 100644
--- a/cuda-oxide/intrinsics/probes/nwarpid.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/nwarpid.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/packed_atomic_add_bf16x2.ll b/cuda-oxide/intrinsics/probes/packed_atomic_add_bf16x2.ll
index 8ce88380d6..1111cab0e8 100644
--- a/cuda-oxide/intrinsics/probes/packed_atomic_add_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/packed_atomic_add_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/packed_atomic_add_f16x2.ll b/cuda-oxide/intrinsics/probes/packed_atomic_add_f16x2.ll
index d45b9003c9..0ec238e0da 100644
--- a/cuda-oxide/intrinsics/probes/packed_atomic_add_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/packed_atomic_add_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/pmevent.ll b/cuda-oxide/intrinsics/probes/pmevent.ll
index 661b4b1ae9..ff9af628e0 100644
--- a/cuda-oxide/intrinsics/probes/pmevent.ll
+++ b/cuda-oxide/intrinsics/probes/pmevent.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prefetch_tma_descriptor.ll b/cuda-oxide/intrinsics/probes/prefetch_tma_descriptor.ll
index 06c1ccb4b4..dc36ea55be 100644
--- a/cuda-oxide/intrinsics/probes/prefetch_tma_descriptor.ll
+++ b/cuda-oxide/intrinsics/probes/prefetch_tma_descriptor.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt.ll b/cuda-oxide/intrinsics/probes/prmt.ll
index 8d363b04c1..e7d238a19f 100644
--- a/cuda-oxide/intrinsics/probes/prmt.ll
+++ b/cuda-oxide/intrinsics/probes/prmt.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt_b4e.ll b/cuda-oxide/intrinsics/probes/prmt_b4e.ll
index fdd4f96867..a1218e54b3 100644
--- a/cuda-oxide/intrinsics/probes/prmt_b4e.ll
+++ b/cuda-oxide/intrinsics/probes/prmt_b4e.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt_ecl.ll b/cuda-oxide/intrinsics/probes/prmt_ecl.ll
index d9f3640d85..bbb2e73e77 100644
--- a/cuda-oxide/intrinsics/probes/prmt_ecl.ll
+++ b/cuda-oxide/intrinsics/probes/prmt_ecl.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt_ecr.ll b/cuda-oxide/intrinsics/probes/prmt_ecr.ll
index 8f22bef597..48b9949bd5 100644
--- a/cuda-oxide/intrinsics/probes/prmt_ecr.ll
+++ b/cuda-oxide/intrinsics/probes/prmt_ecr.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt_f4e.ll b/cuda-oxide/intrinsics/probes/prmt_f4e.ll
index 45e26de539..bbcd5d8d54 100644
--- a/cuda-oxide/intrinsics/probes/prmt_f4e.ll
+++ b/cuda-oxide/intrinsics/probes/prmt_f4e.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt_rc16.ll b/cuda-oxide/intrinsics/probes/prmt_rc16.ll
index 16b565c1d1..aae7467217 100644
--- a/cuda-oxide/intrinsics/probes/prmt_rc16.ll
+++ b/cuda-oxide/intrinsics/probes/prmt_rc16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/prmt_rc8.ll b/cuda-oxide/intrinsics/probes/prmt_rc8.ll
index 1fde8d5fe8..8830088f2e 100644
--- a/cuda-oxide/intrinsics/probes/prmt_rc8.ll
+++ b/cuda-oxide/intrinsics/probes/prmt_rc8.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f32.ll
index fef0dba0bd..cb9e316c01 100644
--- a/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f64.ll b/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f64.ll
index 105ef4a2e5..ad7ad7aaef 100644
--- a/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_approx_ftz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rm_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rm_f32.ll
index b441204b53..e16fed6685 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rm_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rm_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rm_f64.ll b/cuda-oxide/intrinsics/probes/rcp_rm_f64.ll
index cf4b693faa..e52fc61d12 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rm_f64.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rm_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rm_ftz_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rm_ftz_f32.ll
index d389565526..818212973b 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rm_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rm_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rn_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rn_f32.ll
index d6a387a560..e0357d9a35 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rn_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rn_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rn_f64.ll b/cuda-oxide/intrinsics/probes/rcp_rn_f64.ll
index da02ad2519..f1cf96b1e4 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rn_f64.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rn_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rn_ftz_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rn_ftz_f32.ll
index 4d25aa11fe..d534303ff7 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rn_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rn_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rp_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rp_f32.ll
index bebb01b8b7..d116a4b0fd 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rp_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rp_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rp_f64.ll b/cuda-oxide/intrinsics/probes/rcp_rp_f64.ll
index 61ff35afaf..4e6231b9b1 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rp_f64.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rp_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rp_ftz_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rp_ftz_f32.ll
index 0263af7acd..91617991d7 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rp_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rp_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rz_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rz_f32.ll
index 923614eafa..42337793ae 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rz_f64.ll b/cuda-oxide/intrinsics/probes/rcp_rz_f64.ll
index 607567956e..211d1bb57f 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rcp_rz_ftz_f32.ll b/cuda-oxide/intrinsics/probes/rcp_rz_ftz_f32.ll
index 52af228886..ce89a31458 100644
--- a/cuda-oxide/intrinsics/probes/rcp_rz_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rcp_rz_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_add.ll b/cuda-oxide/intrinsics/probes/redux_sync_add.ll
index 6fa409fedf..0c2e753bb3 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_add.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_add.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_and.ll b/cuda-oxide/intrinsics/probes/redux_sync_and.ll
index f636eaa949..b8c214ab64 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_and.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_and.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_max_abs_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_max_abs_f32.ll
index ed3d2b2b6b..71ed83b2fc 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_max_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_max_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_max_abs_nan_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_max_abs_nan_f32.ll
index 86c46f8c1b..26e5f081cb 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_max_abs_nan_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_max_abs_nan_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_max_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_max_f32.ll
index 7df4271426..fbc23a8b97 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_max_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_max_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_max_i32.ll b/cuda-oxide/intrinsics/probes/redux_sync_max_i32.ll
index ffe9bdf596..119729a5fb 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_max_i32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_max_i32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_max_nan_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_max_nan_f32.ll
index 1e8d732997..602124e3a7 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_max_nan_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_max_nan_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_max_u32.ll b/cuda-oxide/intrinsics/probes/redux_sync_max_u32.ll
index 94c0e36224..26e1140870 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_max_u32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_max_u32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_min_abs_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_min_abs_f32.ll
index acd58ca578..e9770680f2 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_min_abs_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_min_abs_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_min_abs_nan_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_min_abs_nan_f32.ll
index 024477177d..0ca86b7845 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_min_abs_nan_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_min_abs_nan_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_min_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_min_f32.ll
index dadb7ee5ba..f96e6ddf2f 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_min_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_min_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_min_i32.ll b/cuda-oxide/intrinsics/probes/redux_sync_min_i32.ll
index 1b68195374..ffe5fd9168 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_min_i32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_min_i32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_min_nan_f32.ll b/cuda-oxide/intrinsics/probes/redux_sync_min_nan_f32.ll
index 7acfed2132..ee0fc0d5a4 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_min_nan_f32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_min_nan_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_min_u32.ll b/cuda-oxide/intrinsics/probes/redux_sync_min_u32.ll
index 7ff8b78ae2..457e80576a 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_min_u32.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_min_u32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_or.ll b/cuda-oxide/intrinsics/probes/redux_sync_or.ll
index e89a02af0a..ed161fd7e1 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_or.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_or.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/redux_sync_xor.ll b/cuda-oxide/intrinsics/probes/redux_sync_xor.ll
index 84b947e298..c03c50cf02 100644
--- a/cuda-oxide/intrinsics/probes/redux_sync_xor.ll
+++ b/cuda-oxide/intrinsics/probes/redux_sync_xor.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rsqrt_approx_f32.ll b/cuda-oxide/intrinsics/probes/rsqrt_approx_f32.ll
index de5e025417..e4d7228376 100644
--- a/cuda-oxide/intrinsics/probes/rsqrt_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rsqrt_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rsqrt_approx_f64.ll b/cuda-oxide/intrinsics/probes/rsqrt_approx_f64.ll
index c381bb0c12..4f48c804f1 100644
--- a/cuda-oxide/intrinsics/probes/rsqrt_approx_f64.ll
+++ b/cuda-oxide/intrinsics/probes/rsqrt_approx_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/rsqrt_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/rsqrt_approx_ftz_f32.ll
index d798d5a19f..828ad30b00 100644
--- a/cuda-oxide/intrinsics/probes/rsqrt_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/rsqrt_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/setmaxnreg_dec.ll b/cuda-oxide/intrinsics/probes/setmaxnreg_dec.ll
index af37b39561..d2cba3359d 100644
--- a/cuda-oxide/intrinsics/probes/setmaxnreg_dec.ll
+++ b/cuda-oxide/intrinsics/probes/setmaxnreg_dec.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/setmaxnreg_inc.ll b/cuda-oxide/intrinsics/probes/setmaxnreg_inc.ll
index 6a5a12f145..e1953f3cfa 100644
--- a/cuda-oxide/intrinsics/probes/setmaxnreg_inc.ll
+++ b/cuda-oxide/intrinsics/probes/setmaxnreg_inc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_down_f32_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_down_f32_sync.ll
index f0cc07b049..db2087224f 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_down_f32_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_down_f32_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_down_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_down_sync.ll
index 0c190bb994..02a95a332e 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_down_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_down_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_down_u64_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_down_u64_sync.ll
index 72e66c3ce4..6b0e005a30 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_down_u64_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_down_u64_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_f32_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_f32_sync.ll
index 4e49b2feb7..c627065725 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_f32_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_f32_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_sync.ll
index 42ecc23be7..e6f72c28ae 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_u64_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_u64_sync.ll
index fe100ff054..2a7a9a7a70 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_u64_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_u64_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_up_f32_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_up_f32_sync.ll
index 38729f4ba6..c14e1bc459 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_up_f32_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_up_f32_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_up_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_up_sync.ll
index 9e09f6f83a..bb43fd0b6d 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_up_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_up_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_up_u64_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_up_u64_sync.ll
index 601928804b..13be404b92 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_up_u64_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_up_u64_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_xor_f32_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_xor_f32_sync.ll
index 95da54f491..37030a5272 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_xor_f32_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_xor_f32_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_xor_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_xor_sync.ll
index 4efa1e64ca..b25cc907c1 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_xor_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_xor_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/shuffle_xor_u64_sync.ll b/cuda-oxide/intrinsics/probes/shuffle_xor_u64_sync.ll
index e6862cc95c..46b9532127 100644
--- a/cuda-oxide/intrinsics/probes/shuffle_xor_u64_sync.ll
+++ b/cuda-oxide/intrinsics/probes/shuffle_xor_u64_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sin_approx_f32.ll b/cuda-oxide/intrinsics/probes/sin_approx_f32.ll
index cb89352f47..56238c93d7 100644
--- a/cuda-oxide/intrinsics/probes/sin_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sin_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sin_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/sin_approx_ftz_f32.ll
index b1fd971f5c..e46f7410da 100644
--- a/cuda-oxide/intrinsics/probes/sin_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sin_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/smid.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/smid.lib_nvvm.ll
index a73644934d..4d5c293ef1 100644
--- a/cuda-oxide/intrinsics/probes/smid.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/smid.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/smid.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/smid.llvm_nvptx.ll
index eb1d182e2d..597b0dcc44 100644
--- a/cuda-oxide/intrinsics/probes/smid.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/smid.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_approx_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_approx_f32.ll
index 72ff6b20c3..66cae7bf2d 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_approx_ftz_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_approx_ftz_f32.ll
index 280aab1b93..f30d3a602f 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_approx_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_approx_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rm_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rm_f32.ll
index 8f985c9cd7..e4f66f1f98 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rm_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rm_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rm_f64.ll b/cuda-oxide/intrinsics/probes/sqrt_rm_f64.ll
index a83ca9bc74..2ffca499c0 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rm_f64.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rm_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rm_ftz_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rm_ftz_f32.ll
index 15e453699e..ae698906d1 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rm_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rm_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rn_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rn_f32.ll
index 31fee94478..a0f3e67b69 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rn_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rn_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rn_f64.ll b/cuda-oxide/intrinsics/probes/sqrt_rn_f64.ll
index e9a3f1ffbe..84f69a4c2b 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rn_f64.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rn_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rn_ftz_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rn_ftz_f32.ll
index 3777543af4..2d18411bf8 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rn_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rn_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rp_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rp_f32.ll
index 82037c4860..e582bd4225 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rp_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rp_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rp_f64.ll b/cuda-oxide/intrinsics/probes/sqrt_rp_f64.ll
index 9604d98682..e36ad71f3d 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rp_f64.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rp_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rp_ftz_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rp_ftz_f32.ll
index 75ac3ae913..d77884c98c 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rp_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rp_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rz_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rz_f32.ll
index 367661169b..08308e002c 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rz_f64.ll b/cuda-oxide/intrinsics/probes/sqrt_rz_f64.ll
index b1b24314cc..abe907ac33 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rz_f64.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rz_f64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sqrt_rz_ftz_f32.ll b/cuda-oxide/intrinsics/probes/sqrt_rz_ftz_f32.ll
index 01548379a0..e6498b0c18 100644
--- a/cuda-oxide/intrinsics/probes/sqrt_rz_ftz_f32.ll
+++ b/cuda-oxide/intrinsics/probes/sqrt_rz_ftz_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_b16.ll b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_b16.ll
index 0d8744f163..8b1fbcef6e 100644
--- a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_b16.ll
+++ b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_trans_b16.ll b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_trans_b16.ll
index cd059aa51f..32b43dd8c2 100644
--- a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_trans_b16.ll
+++ b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x2_trans_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_b16.ll b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_b16.ll
index 972f13e200..5db2851f97 100644
--- a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_b16.ll
+++ b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_trans_b16.ll b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_trans_b16.ll
index 79566adecb..4f7d355392 100644
--- a/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_trans_b16.ll
+++ b/cuda-oxide/intrinsics/probes/stmatrix_m8n8_x4_trans_b16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sub_bf16x2.ll b/cuda-oxide/intrinsics/probes/sub_bf16x2.ll
index 02007ab138..8970e79012 100644
--- a/cuda-oxide/intrinsics/probes/sub_bf16x2.ll
+++ b/cuda-oxide/intrinsics/probes/sub_bf16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sub_f16x2.ll b/cuda-oxide/intrinsics/probes/sub_f16x2.ll
index ee084d0d9d..d77114a98d 100644
--- a/cuda-oxide/intrinsics/probes/sub_f16x2.ll
+++ b/cuda-oxide/intrinsics/probes/sub_f16x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sub_f32x2.ll b/cuda-oxide/intrinsics/probes/sub_f32x2.ll
index c4dbe9a3d3..686258e982 100644
--- a/cuda-oxide/intrinsics/probes/sub_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/sub_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sub_ftz_f32x2.ll b/cuda-oxide/intrinsics/probes/sub_ftz_f32x2.ll
index 9043eb52e7..a402a2c3f7 100644
--- a/cuda-oxide/intrinsics/probes/sub_ftz_f32x2.ll
+++ b/cuda-oxide/intrinsics/probes/sub_ftz_f32x2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sync_mask.ll b/cuda-oxide/intrinsics/probes/sync_mask.ll
index ee16870a76..683b1be50a 100644
--- a/cuda-oxide/intrinsics/probes/sync_mask.ll
+++ b/cuda-oxide/intrinsics/probes/sync_mask.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/sync_threads.ll b/cuda-oxide/intrinsics/probes/sync_threads.ll
index a8130be45f..bfa844406b 100644
--- a/cuda-oxide/intrinsics/probes/sync_threads.ll
+++ b/cuda-oxide/intrinsics/probes/sync_threads.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tanh_approx_f32.ll b/cuda-oxide/intrinsics/probes/tanh_approx_f32.ll
index e3e34ae1df..5f0d7e4493 100644
--- a/cuda-oxide/intrinsics/probes/tanh_approx_f32.ll
+++ b/cuda-oxide/intrinsics/probes/tanh_approx_f32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_alloc.ll b/cuda-oxide/intrinsics/probes/tcgen05_alloc.ll
index 6acfe45a1c..b823ac413b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_alloc.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_alloc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_alloc_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_alloc_cg2.ll
index dc0cbca220..dfb302cecb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_alloc_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_alloc_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_commit.ll b/cuda-oxide/intrinsics/probes/tcgen05_commit.ll
index beae2ad8f3..779a70d70e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_commit.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_commit.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_commit_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_commit_cg2.ll
index 023e507d30..cb0f8096d4 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_commit_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_commit_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast.ll b/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast.ll
index 699a249e90..2e30ccc923 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast_cg2.ll
index 4772d30167..f92286e1f9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_commit_multicast_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster.ll b/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster.ll
index acb1289973..59af03b61b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster_cg2.ll
index a1433b90ed..df4b9c0626 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_commit_shared_cluster_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b.ll
index 874d863547..a56fb45429 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64.ll
index f4e3048836..31a987f4e0 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64_cg2.ll
index 718a763a12..1b5002737d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b4x16_p64_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32.ll
index 2d06edfdf3..b2ff7cd266 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32_cg2.ll
index 035c14ab55..5a92491c7c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_b6x16_p32_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_cg2.ll
index 57f39a88a9..40c0f29f78 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x128b_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64.ll
index be74fe61f7..12f23ad13d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64_cg2.ll
index c011f0964b..09e05d6ef3 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b4x16_p64_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32.ll
index d129867b93..431b81da7a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32_cg2.ll
index 91b0ad59d5..09774f7173 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_128x256b_b6x16_p32_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4.ll
index 388a3a637d..e6128ec64b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64.ll
index e3d340db5d..a729bc01d9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2.ll
index 544639637e..e63edfab81 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b4x16_p64_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32.ll
index b88e27ade7..dca98ba632 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2.ll
index 0f61ebfa29..3f05d3a6e2 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_b6x16_p32_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_cg2.ll
index 75716157c3..c09e87a586 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_32x128b_warpx4_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b.ll
index d0477d394b..8d4d634cbb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64.ll
index bc8bcb758c..822e77604e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64_cg2.ll
index b5008cc128..1d376316dc 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b4x16_p64_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32.ll
index 2404cfa452..272e8ef0e1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32_cg2.ll
index 6346a9b18c..bf72fbf488 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_b6x16_p32_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_cg2.ll
index 0823b74275..5b4bd94c8d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_4x256b_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23.ll
index 851638d6ee..7837962693 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64.ll
index 6154c24fd5..0c8ef2cfd3 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2.ll
index 9c81cd3854..a78f8aa6d6 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b4x16_p64_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32.ll
index e6232c72ef..a0b7e81098 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2.ll
index 04eb00bd45..02a4d9562c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_b6x16_p32_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_cg2.ll
index 913e0c9653..14d660a9cc 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_01_23_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13.ll
index cd6b4548c8..8656971789 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64.ll
index ca2f614a30..7bd6d95497 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2.ll
index ea163ebf31..d646da8df8 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b4x16_p64_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32.ll
index 900231f06d..82a1b27449 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2.ll
index ee2c9aa47a..69ca6f3776 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_b6x16_p32_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_cg2.ll
index 4b53f40f9a..82d525ada3 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_64x128b_warpx2_02_13_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem.ll
index f2927ff61e..21adeb6aec 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem_cg2.ll
index 2c6d574821..f335cec4d1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_cp_smem_to_tmem_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_dealloc.ll b/cuda-oxide/intrinsics/probes/tcgen05_dealloc.ll
index 7bae53ecbc..34170acc77 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_dealloc.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_dealloc.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_dealloc_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_dealloc_cg2.ll
index 9c5c87deee..6c4b3c8e84 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_dealloc_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_dealloc_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_fence_after_thread_sync.ll b/cuda-oxide/intrinsics/probes/tcgen05_fence_after_thread_sync.ll
index 4d99f51ecf..463b96a822 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_fence_after_thread_sync.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_fence_after_thread_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_fence_before_thread_sync.ll b/cuda-oxide/intrinsics/probes/tcgen05_fence_before_thread_sync.ll
index 2e85368df9..7eb0ca8654 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_fence_before_thread_sync.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_fence_before_thread_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_pack16.ll
index 66eea7805c..5ae9bd0f15 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_raw.ll
index ebcd6d5324..586087f6a9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_pack16.ll
index 122e4ea3af..fed4990a1b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_raw.ll
index ed09506f35..647f31d933 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_pack16.ll
index 2c16e905b2..650cea233f 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_raw.ll
index cb6e7fafc3..3c1b0e92fd 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_pack16.ll
index 38733d57e4..910354cb3f 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_raw.ll
index 65fd5cc7b5..eca0f558cb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_pack16.ll
index 0676b6a0f6..e9bbbe0942 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_raw.ll
index e40101f924..0231c8cee7 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_pack16.ll
index dedb0a0aa8..18d52449a3 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_raw.ll
index cc8df14d6f..52c3931b6d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_pack16.ll
index 3fd33fb9fb..5c14240d1c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_raw.ll
index ed89859528..d5780d2b58 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x128b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_pure.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_pure.ll
index a54c69118f..e473742565 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_pure.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_pure.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_pack16.ll
index ea56b1fcc8..4446a354ea 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_raw.ll
index 9a97fcdb32..b45bcd8f64 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_pack16.ll
index e0edac159d..bfd93db9ea 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_raw.ll
index e900efdf2c..61e93d7b86 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_pack16.ll
index 77aadce51d..b970b32f99 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_raw.ll
index 8c31fbdd5f..5e6ec4657f 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_pack16.ll
index 9e62b6aa48..f869e87f30 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_raw.ll
index 62e4a4c350..197489f82c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_pack16.ll
index 26d54c0e9b..070622fe59 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_raw.ll
index 89f8e40fb3..03fb61e7fb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pack16.ll
index 6a77ff0120..4a51274d3b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pure.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pure.ll
index d50558f4de..90cb0d5fe4 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pure.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_pure.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_raw.ll
index 40dbcdecff..2a94f9813d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x256b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_pack16.ll
index 4ffa096247..4e2ab19a27 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_raw.ll
index 80e486edc9..b034aee05b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x128_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_pack16.ll
index 3bc438421d..5fe0a8efed 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_raw.ll
index b525a53143..83cac7b1e1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_pack16.ll
index ea91a76523..71aa4fc632 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_raw.ll
index 00da06b059..43fac69148 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_pack16.ll
index 4c8c9985a6..22f0043a56 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_raw.ll
index 4b3daade89..261adbdbc9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_pack16.ll
index d11833737e..e5f9e24f90 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_raw.ll
index 0c7fcf9f74..02be0e8cad 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_pack16.ll
index 03733b0c51..e74d6957ac 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_raw.ll
index 43a58a1463..371d6f78b0 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_pack16.ll
index cbbf017da2..71c6cc8313 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_raw.ll
index d9a4c1e9b7..722bbb9938 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_pack16.ll
index dd2114bcc3..68d013a0dc 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_raw.ll
index be835e1829..d9db4b69fd 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x32bx2_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_pack16.ll
index 1f9edb56a4..578be1d4dd 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_raw.ll
index f18bbd320d..730e54dbfb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x128_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_pack16.ll
index e29c844fda..71c61e2deb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_raw.ll
index 624bc96f51..183b1767e6 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_pack16.ll
index bda45b63ea..bb25fecc3c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_raw.ll
index 8f179c43fb..e527fb2524 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_pack16.ll
index 0c0110ad7c..15dcf1f873 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_raw.ll
index 8b9844f114..f59cd79143 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_pack16.ll
index 5f44b02772..8947077001 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_raw.ll
index d2416f0b00..23f0103b48 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_pack16.ll
index 93e718a265..e2003259a6 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_raw.ll
index f1b7f0bfdb..83f8e8f89c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_pack16.ll
index 911d27c9de..01c6816cfb 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_raw.ll
index b2da613b08..1f54053c1c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_pack16.ll
index f803c14f8c..8c24485dba 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_raw.ll
index 41b56f833c..4165c98728 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_16x64b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_pack16.ll
index cf894651a3..dcf3eb1268 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_raw.ll
index eb6d10e4ee..bd0569970f 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x128_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_pack16.ll
index 1de6de1955..830163dda2 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_raw.ll
index 2a64bffd5b..41bf253cc1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_pack16.ll
index 2d5f8d1a89..fa2725a12d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_raw.ll
index c00badb7f2..89ca9a137d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_pack16.ll
index baaf6a82bd..f290ab329c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_raw.ll
index 73e2290d58..f1bbfeef13 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_pack16.ll
index a32e916295..1f4b9e89b4 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_raw.ll
index ea7f80426f..e46f0ca74e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_pack16.ll
index eb7afc20a7..a7d62aaf59 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_raw.ll
index 114b18dc87..c94cd41efc 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_pack16.ll
index 78cfc8971e..dbf59bafdd 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_raw.ll
index 12f42ca913..6dead57bca 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_pack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_pack16.ll
index 8695c8b553..9ea197a58a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_pack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_pack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_raw.ll
index 690bb364d6..74df38e7c9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_ld_32x32b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_load_wait.ll b/cuda-oxide/intrinsics/probes/tcgen05_load_wait.ll
index 2bdc2cd00e..d3fd44cf7b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_load_wait.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_load_wait.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m1.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m1.ll
index 3d86f3ec30..2c1773e38b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m3.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m3.ll
index 35aa6aad96..e975159ebf 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_e3m2.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_e3m2.ll
index efddea9aba..54fbc8fca1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_e4m3.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_e4m3.ll
index a0d6aea99c..a160caea4e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_e5m2.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_e5m2.ll
index 4b442a42ea..fa6921722a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_f16.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_f16.ll
index ea9cbec878..8de63e138d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_f16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_f16_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_f16_cg2.ll
index 2ff1642e13..2f264a9b12 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_f16_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_f16_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_shared.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_shared.ll
index bd1a07c0b6..0f32015fb1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_shared.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_shared.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_shared.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_shared.ll
index 8e9bf92012..35f7bc7167 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_shared.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_shared.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor.ll
index 6cf4f5d3d3..9117cdfd9e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor_ashift.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor_ashift.ll
index 7c7e82d3aa..5facb4658a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor_ashift.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_sp_tensor_ashift.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor.ll
index 16f6ee0987..748f593bc6 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor_ashift.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor_ashift.ll
index 7c81061cd3..e3a2ddf422 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor_ashift.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_tensor_ashift.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_bf16.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_bf16.ll
index 7bdec871ab..6bb3038653 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_bf16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_bf16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m1.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m1.ll
index 1926738c37..aedbde911b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m1.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m1.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m3.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m3.ll
index 6c8f8d352b..c9211ebf93 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m3.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e2m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e3m2.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e3m2.ll
index edc85afaa4..ff4d689ed2 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e3m2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e3m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e4m3.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e4m3.ll
index d9b5d3ed36..6119c1b78a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e4m3.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e4m3.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e5m2.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e5m2.ll
index 1e8ae5c0fc..3587b55966 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e5m2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_e5m2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_f16.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_f16.ll
index f1e5272d20..3d1513573e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_f16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_f16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared.ll
index 06651cf6d4..db6f697ace 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared_zero_col_mask.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared_zero_col_mask.ll
index 12b9e3e785..94ffd5f6aa 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared_zero_col_mask.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_shared_zero_col_mask.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared.ll
index 6ac3fab437..e9445ec605 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared_zero_col_mask.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared_zero_col_mask.ll
index bdfd06214a..e6d9cdf552 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared_zero_col_mask.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_shared_zero_col_mask.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor.ll
index fda46dbb86..13650f29e6 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor_zero_col_mask.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor_zero_col_mask.ll
index e477013ea4..9881f97c0a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor_zero_col_mask.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_sp_tensor_zero_col_mask.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor.ll
index 8c0f0911a5..981f53b0f8 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor_zero_col_mask.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor_zero_col_mask.ll
index d0a1478f1e..e80ecc14d9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor_zero_col_mask.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tensor_zero_col_mask.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tf32.ll b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tf32.ll
index 1ef72f1863..a341605a9e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tf32.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_mma_ws_tf32.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit.ll b/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit.ll
index e4d3b3cb80..61a4985f62 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit_cg2.ll
index b018f24a45..405b42e46a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_relinquish_alloc_permit_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_shift_down.ll b/cuda-oxide/intrinsics/probes/tcgen05_shift_down.ll
index a21d3b771b..b6cd43c4ed 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_shift_down.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_shift_down.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_shift_down_cg2.ll b/cuda-oxide/intrinsics/probes/tcgen05_shift_down_cg2.ll
index 10cb9cc4ae..edf0f0f845 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_shift_down_cg2.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_shift_down_cg2.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_raw.ll
index 7c429f8205..ba656768e3 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_unpack16.ll
index a4d28ec63b..325dcf58ec 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x16_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_raw.ll
index 04cea28f4e..06939a001b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_unpack16.ll
index 89567ebd34..fe2ce75b77 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x1_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_raw.ll
index 6dc84f6ecb..3f7835672d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_unpack16.ll
index 3aa981e65b..9ce2e04316 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x2_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_raw.ll
index 2ab4eca27c..e0399b1b80 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_unpack16.ll
index e8f73ec97e..ea382a3e4c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x32_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_raw.ll
index 6991738971..eb14712d1a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_unpack16.ll
index 95f5fc3501..cef55806f6 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x4_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_raw.ll
index b405eb8847..cce1d76678 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_unpack16.ll
index 9214e24793..ae71ecb02d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x64_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_raw.ll
index d6824be5c9..1309cf9a45 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_unpack16.ll
index 9a193796f4..561ad2b226 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x128b_x8_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_raw.ll
index fda79237c9..4a7a422acf 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_unpack16.ll
index 8f9d7122b3..102e728e71 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x16_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_raw.ll
index fdf8be91d4..77e2d52b11 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_unpack16.ll
index 7e9d57b1d4..2d6e832e41 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x1_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_raw.ll
index 51f68b428f..8f76bfac3b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_unpack16.ll
index 2a22da8dfb..8ed6bcc5aa 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x2_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_raw.ll
index 78d13c1601..c3d4ae1412 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_unpack16.ll
index 4324fd6448..88b8146e7f 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x32_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_raw.ll
index a02f312eac..e7c2de1788 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_unpack16.ll
index 445d595681..734f56972d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x4_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_raw.ll
index 179fe1d8ce..9368684e3e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_unpack16.ll
index 7f2e27bb63..cd978dbea9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x256b_x8_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_raw.ll
index 94ba07bf66..b452b20c0d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_unpack16.ll
index 318fa8dfef..291d882495 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x128_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_raw.ll
index bfc9cad2b2..334d590622 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_unpack16.ll
index 68278293aa..a1cc78d36b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x16_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_raw.ll
index dd0ebe55da..3fc331675c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_unpack16.ll
index c173bd426c..351a96ea42 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x1_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_raw.ll
index 64704b9451..33179a9019 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_unpack16.ll
index 6c21d82c47..c9e00cdf2e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x2_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_raw.ll
index c33a5a2262..8969ab99f5 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_unpack16.ll
index 78e1a39bb5..13ae0c1322 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x32_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_raw.ll
index 96bec07f91..5615fdc594 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_unpack16.ll
index d613c5f820..ccc1986b6d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x4_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_raw.ll
index 4095f9f91f..76ba355871 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_unpack16.ll
index f69630077d..93ce5bfb45 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x64_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_raw.ll
index 99b477f92b..1d98697b7d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_unpack16.ll
index b88ab5dfe7..6ec802a078 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x32bx2_x8_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_raw.ll
index 230955bafa..54408310a9 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_unpack16.ll
index 6e9e9b3ee1..fce5aab34f 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x128_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_raw.ll
index 19baa5f06f..579190ce50 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_unpack16.ll
index 0133bde57f..945c41b3a1 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x16_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_raw.ll
index e4dcd66474..6c3d0cbc2b 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_unpack16.ll
index 6c19cdad41..ebe4861d94 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x1_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_raw.ll
index ee58d4837d..cc93a7369e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_unpack16.ll
index 8c2b89be5b..8d9474b696 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x2_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_raw.ll
index 97bac36f86..f5a16d9133 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_unpack16.ll
index 41b64f612c..693fc0b853 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x32_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_raw.ll
index 903a071085..2c73baa378 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_unpack16.ll
index dd2c2a4782..5d0cf33455 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x4_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_raw.ll
index f14f075df1..4f3dbb60ef 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_unpack16.ll
index d93185a9d6..57ce4c0886 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x64_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_raw.ll
index 5fe775d3cd..5e85f3a09d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_unpack16.ll
index a544ecb655..9b709f422e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_16x64b_x8_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_raw.ll
index 60a076fe83..5a9c9373e0 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_unpack16.ll
index 545930c809..bee9bffa8c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x128_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_raw.ll
index 827771e3d5..8c77ac107a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_unpack16.ll
index 1721568cf6..c8cc9d74ed 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x16_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_raw.ll
index 15d598cec6..333f2b5205 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_unpack16.ll
index a698bdc75d..ac0554d50e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x1_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_raw.ll
index ecedbf8056..d9eb449d59 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_unpack16.ll
index 6a83680da1..610f0af819 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x2_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_raw.ll
index beea9becc1..33a2ee4102 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_unpack16.ll
index 9b42f9b0a4..548bd59e8d 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x32_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_raw.ll
index ccec22823a..baad23fe6a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_unpack16.ll
index c8b9ddc19c..98d7bbbb7e 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x4_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_raw.ll
index b717e71d43..0ec73445c8 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_unpack16.ll
index 25ac6dd11b..5a3eb0984c 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x64_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_raw.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_raw.ll
index 97b923d8a6..bca6fc8d35 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_raw.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_raw.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_unpack16.ll b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_unpack16.ll
index b810bab3a5..b811beb1f0 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_unpack16.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_st_32x32b_x8_unpack16.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tcgen05_store_wait.ll b/cuda-oxide/intrinsics/probes/tcgen05_store_wait.ll
index 7726764af4..43185ffd6a 100644
--- a/cuda-oxide/intrinsics/probes/tcgen05_store_wait.ll
+++ b/cuda-oxide/intrinsics/probes/tcgen05_store_wait.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_box_dim.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_box_dim.ll
index 383215df50..d0f0f31dd9 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_box_dim.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_box_dim.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_element_stride.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_element_stride.ll
index aa4f704382..cb489ffa7c 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_element_stride.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_element_stride.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_element_type.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_element_type.ll
index 10bd454b1a..b65e7f27aa 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_element_type.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_element_type.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_fill_mode.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_fill_mode.ll
index 4a4137fb54..02e4af589f 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_fill_mode.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_fill_mode.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_global_address.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_global_address.ll
index 4ae2808853..8c289fca96 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_global_address.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_global_address.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_global_dim.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_global_dim.ll
index 61aa6c25e2..1e1f8ae263 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_global_dim.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_global_dim.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_global_stride.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_global_stride.ll
index c0758977bb..aee07673e4 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_global_stride.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_global_stride.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_interleave_layout.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_interleave_layout.ll
index cd114237d2..ce6ff316b2 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_interleave_layout.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_interleave_layout.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_rank.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_rank.ll
index fa0adf79fc..e335b2b467 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_rank.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_rank.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_atomicity.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_atomicity.ll
index dc59ddd5bd..afb08c6673 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_atomicity.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_atomicity.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_mode.ll b/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_mode.ll
index 79573631db..d63b34a14c 100644
--- a/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_mode.ll
+++ b/cuda-oxide/intrinsics/probes/tensormap_replace_swizzle_mode.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/thread_idx_x.ll b/cuda-oxide/intrinsics/probes/thread_idx_x.ll
index 70f54c724c..406589b9ea 100644
--- a/cuda-oxide/intrinsics/probes/thread_idx_x.ll
+++ b/cuda-oxide/intrinsics/probes/thread_idx_x.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/thread_idx_y.ll b/cuda-oxide/intrinsics/probes/thread_idx_y.ll
index d8623f0ce3..df4ebf993a 100644
--- a/cuda-oxide/intrinsics/probes/thread_idx_y.ll
+++ b/cuda-oxide/intrinsics/probes/thread_idx_y.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/thread_idx_z.ll b/cuda-oxide/intrinsics/probes/thread_idx_z.ll
index 94614f0fda..04e6cb9ae7 100644
--- a/cuda-oxide/intrinsics/probes/thread_idx_z.ll
+++ b/cuda-oxide/intrinsics/probes/thread_idx_z.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/threadfence.ll b/cuda-oxide/intrinsics/probes/threadfence.ll
index 4dcb217402..3e40aeed62 100644
--- a/cuda-oxide/intrinsics/probes/threadfence.ll
+++ b/cuda-oxide/intrinsics/probes/threadfence.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/threadfence_block.ll b/cuda-oxide/intrinsics/probes/threadfence_block.ll
index cbb9417b54..c75307a800 100644
--- a/cuda-oxide/intrinsics/probes/threadfence_block.ll
+++ b/cuda-oxide/intrinsics/probes/threadfence_block.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/threadfence_system.ll b/cuda-oxide/intrinsics/probes/threadfence_system.ll
index c6aa00abc1..826e9530a3 100644
--- a/cuda-oxide/intrinsics/probes/threadfence_system.ll
+++ b/cuda-oxide/intrinsics/probes/threadfence_system.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/total_smem_size.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/total_smem_size.lib_nvvm.ll
index 56ddcc2e84..4d92137b25 100644
--- a/cuda-oxide/intrinsics/probes/total_smem_size.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/total_smem_size.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/total_smem_size.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/total_smem_size.llvm_nvptx.ll
index 8ac62d912e..5707a1d084 100644
--- a/cuda-oxide/intrinsics/probes/total_smem_size.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/total_smem_size.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/trap.ll b/cuda-oxide/intrinsics/probes/trap.ll
index 50e301518a..6da7e92ab2 100644
--- a/cuda-oxide/intrinsics/probes/trap.ll
+++ b/cuda-oxide/intrinsics/probes/trap.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/uni_sync.ll b/cuda-oxide/intrinsics/probes/uni_sync.ll
index c90bea50bf..a33eef123c 100644
--- a/cuda-oxide/intrinsics/probes/uni_sync.ll
+++ b/cuda-oxide/intrinsics/probes/uni_sync.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/warpid.lib_nvvm.ll b/cuda-oxide/intrinsics/probes/warpid.lib_nvvm.ll
index 647cb3118a..3106842b64 100644
--- a/cuda-oxide/intrinsics/probes/warpid.lib_nvvm.ll
+++ b/cuda-oxide/intrinsics/probes/warpid.lib_nvvm.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/warpid.llvm_nvptx.ll b/cuda-oxide/intrinsics/probes/warpid.llvm_nvptx.ll
index a9453df671..5a050dcd77 100644
--- a/cuda-oxide/intrinsics/probes/warpid.llvm_nvptx.ll
+++ b/cuda-oxide/intrinsics/probes/warpid.llvm_nvptx.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/wgmma_commit_group.ll b/cuda-oxide/intrinsics/probes/wgmma_commit_group.ll
index e6c502f5d5..e9bf779030 100644
--- a/cuda-oxide/intrinsics/probes/wgmma_commit_group.ll
+++ b/cuda-oxide/intrinsics/probes/wgmma_commit_group.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/wgmma_fence.ll b/cuda-oxide/intrinsics/probes/wgmma_fence.ll
index fc90766de2..0f230c57dd 100644
--- a/cuda-oxide/intrinsics/probes/wgmma_fence.ll
+++ b/cuda-oxide/intrinsics/probes/wgmma_fence.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/probes/wgmma_wait_group.ll b/cuda-oxide/intrinsics/probes/wgmma_wait_group.ll
index 8691bd31ae..25ac4f95cc 100644
--- a/cuda-oxide/intrinsics/probes/wgmma_wait_group.ll
+++ b/cuda-oxide/intrinsics/probes/wgmma_wait_group.ll
@@ -1,8 +1,8 @@
 ; SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 ; SPDX-License-Identifier: Apache-2.0
 ; @generated by cuda-intrinsics-gen 0.1.0. DO NOT EDIT.
-; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: 016c9d9b55b84d99d56fe2690dce4b3ba7ded5318890996a7459a142ddb5edf7.
-; LLVM source: 16696adcd119e6ba9cc175207d984d7021211acb.
+; Catalog schema/version: 46/0.1.0; intrinsic ABI: v1; catalog SHA-256: c243e568fc16c9f907c9fd87fbacbe9f823043a9badd8633e43e9faa87adbd10.
+; LLVM source: aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019.
 
 target triple = "nvptx64-nvidia-cuda"
 
diff --git a/cuda-oxide/intrinsics/upstream.lock b/cuda-oxide/intrinsics/upstream.lock
index 8de13fd6ff..e13c5668f9 100644
--- a/cuda-oxide/intrinsics/upstream.lock
+++ b/cuda-oxide/intrinsics/upstream.lock
@@ -2,29 +2,29 @@ schema = 1
 
 [llvm]
 repository = "https://github.com/rust-lang/llvm-project.git"
-revision = "16696adcd119e6ba9cc175207d984d7021211acb"
-provenance = "Public rust-lang/llvm-project source revision used by Rust commit e457a7b0d"
+revision = "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019"
+provenance = "Public rust-lang/llvm-project source revision used by Rust commit 6f5da406b"
 public_output_allowed = true
 
 [rust_toolchain]
-commit_hash = "e457a7b0d326d67b4322ef0d11bd715cfaeda48f"
+commit_hash = "6f5da406b56c2db6607202b63daee73aa8c61297"
 
 [llvm_tblgen]
 name = "llvm-tblgen"
-version_line = "LLVM version 23.1.0"
-sha256 = "199fe46b773c4f5b38f3e365dea2408e0f344ce8873a14c7497549ff1c90a9ba"
+version_line = "LLVM version 23.1.3"
+sha256 = "f9e28cd046b8084efed066634a239076763ac7a7a6bde932b58e2cb439c25564"
 enforce_sha256 = false
 provenance = "Reference artifact built locally from the pinned public LLVM revision; its hash identifies that build, while dump hashes are the reproducibility gate"
-built_from_llvm_revision = "16696adcd119e6ba9cc175207d984d7021211acb"
+built_from_llvm_revision = "aab11ee0ea5ea9a1d462ad0bc1fae9d9bb643019"
 
 [[comparison_tools]]
 name = "llvm-tblgen-22"
-version_line = "Ubuntu LLVM version 22.1.7"
-sha256 = "986d0d5a19e4f44086092adf9a54f13ad0058a60ac2adad7ba156a97eb01b046"
+version_line = "Ubuntu LLVM version 22.1.8"
+sha256 = "dbe41ce85b55b137b846da027f34bdeb260d25b907059a39927eea6aee4cae96"
 enforce_sha256 = false
 provenance = "Ubuntu package binary used only for a byte-for-byte dump comparison; it reproduced the Intrinsics.td dump hash exactly. The NVPTX.td dump uses LLVM 23-only TableGen operators (!sort in Target.td) that pre-23 binaries reject, so no independent binary was available for that dump"
 
 [dumps]
 intrinsics_sha256 = "c0fb05eda32f2fe89ba600d4d2dc882e0bf54d5aa60b5e5f8601eb87a9881408"
 nvptx_sha256 = "492011cd3c20fbd888a2fa05e7eb5d894c0db79270c5bf6dbda0b6648d3c6582"
-normalized_imported_sha256 = "3f54245b231a5b6e19f3225710c3865b7e1a106242cce873250ef5fdc32e532e"
+normalized_imported_sha256 = "54d47b8e281dd953dc95fe16f834ff38164d9809171c9a1af097231b0bc45cab"
diff --git a/cuda-oxide/rust-toolchain.toml b/cuda-oxide/rust-toolchain.toml
index 5cf8cac83f..1f9d49c31d 100644
--- a/cuda-oxide/rust-toolchain.toml
+++ b/cuda-oxide/rust-toolchain.toml
@@ -1,5 +1,5 @@
 [toolchain]
-channel = "nightly-2026-08-28"
+channel = "nightly-2026-10-11"
 components = [
     "rust-src",
     "rustc-dev",